Merge pull request #105 from yaosheng216/patch-24

Update constant.py

.github/workflows/publish.yml

@@ -0,0 +1,24 @@
+name: Publish to Comfy registry
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+      - master
+    paths:
+      - "scepter/workflow/pyproject.toml"
+
+jobs:
+  publish-node:
+    name: Publish Custom Node to registry
+    runs-on: ubuntu-latest
+    # if this is a forked repository. Skipping the workflow.
+    if: github.event.repository.fork == false
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Publish Custom Node
+        uses: Comfy-Org/publish-node-action@main
+        with:
+          ## Add your own personal access token to your Github Repository secrets and reference it here.
+          personal_access_token: ${{ secrets.REGISTRY_ACCESS_TOKEN }}

.gitignore

@@ -9,12 +9,12 @@
 *.bin
 *.idea
 *.csv
+cache
 build
 dist
 dev
 scepter.egg-info
 .readthedocs.yml
-1.9
-#MANIFEST.in
 *resources
 *.ipynb_checkpoints*
+*.vscode

asset/workflow/sdxl_base.json

@@ -0,0 +1,163 @@
+{
+  "last_node_id": 4,
+  "last_link_id": 3,
+  "nodes": [
+    {
+      "id": 2,
+      "type": "PreviewImage",
+      "pos": {
+        "0": 937,
+        "1": 243
+      },
+      "size": {
+        "0": 284.03680419921875,
+        "1": 246
+      },
+      "flags": {
+        "collapsed": false
+      },
+      "order": 2,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 1
+        }
+      ],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "PreviewImage"
+      },
+      "widgets_values": []
+    },
+    {
+      "id": 4,
+      "type": "ParameterNode",
+      "pos": {
+        "0": 0,
+        "1": 244
+      },
+      "size": {
+        "0": 311.9849853515625,
+        "1": 236.4765625
+      },
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            3
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ParameterNode"
+      },
+      "widgets_values": [
+        "ddim",
+        50,
+        5,
+        0.5,
+        "trailing",
+        1024,
+        1024,
+        2024
+      ]
+    },
+    {
+      "id": 1,
+      "type": "ModelNode",
+      "pos": {
+        "0": 430,
+        "1": 243
+      },
+      "size": {
+        "0": 400,
+        "1": 234
+      },
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "parameters",
+          "type": "CONDITIONING",
+          "link": 3,
+          "shape": 7
+        },
+        {
+          "name": "mantras",
+          "type": "CONDITIONING",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "tuners",
+          "type": "CONDITIONING",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "controls",
+          "type": "CONDITIONING",
+          "link": null,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            1
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ModelNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0",
+        "ModelScope",
+        "a cat",
+        ""
+      ]
+    }
+  ],
+  "links": [
+    [
+      1,
+      1,
+      0,
+      2,
+      0,
+      "IMAGE"
+    ],
+    [
+      3,
+      4,
+      0,
+      1,
+      0,
+      "CONDITIONING"
+    ]
+  ],
+  "groups": [],
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 0.9229599817706423,
+      "offset": [
+        128.91466506592707,
+        43.07074577975898
+      ]
+    }
+  },
+  "version": 0.4
+}

asset/workflow/sdxl_base_mantra.json

@@ -0,0 +1,202 @@
+{
+  "last_node_id": 5,
+  "last_link_id": 4,
+  "nodes": [
+    {
+      "id": 2,
+      "type": "PreviewImage",
+      "pos": {
+        "0": 937,
+        "1": 243
+      },
+      "size": {
+        "0": 284.03680419921875,
+        "1": 246
+      },
+      "flags": {
+        "collapsed": false
+      },
+      "order": 3,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 1
+        }
+      ],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "PreviewImage"
+      },
+      "widgets_values": []
+    },
+    {
+      "id": 4,
+      "type": "ParameterNode",
+      "pos": {
+        "0": 0,
+        "1": 244
+      },
+      "size": {
+        "0": 311.9849853515625,
+        "1": 236.4765625
+      },
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            3
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ParameterNode"
+      },
+      "widgets_values": [
+        "ddim",
+        50,
+        5,
+        0.5,
+        "trailing",
+        1024,
+        1024,
+        2024
+      ]
+    },
+    {
+      "id": 1,
+      "type": "ModelNode",
+      "pos": {
+        "0": 430,
+        "1": 243
+      },
+      "size": {
+        "0": 400,
+        "1": 234
+      },
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "parameters",
+          "type": "CONDITIONING",
+          "link": 3,
+          "shape": 7
+        },
+        {
+          "name": "mantras",
+          "type": "CONDITIONING",
+          "link": 4,
+          "shape": 7
+        },
+        {
+          "name": "tuners",
+          "type": "CONDITIONING",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "controls",
+          "type": "CONDITIONING",
+          "link": null,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            1
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ModelNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0",
+        "ModelScope",
+        "a cat",
+        ""
+      ]
+    },
+    {
+      "id": 5,
+      "type": "MantrasNode",
+      "pos": {
+        "0": 3,
+        "1": 547
+      },
+      "size": {
+        "0": 315,
+        "1": 58
+      },
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            4
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "MantrasNode"
+      },
+      "widgets_values": [
+        "Flat 2D Art"
+      ]
+    }
+  ],
+  "links": [
+    [
+      1,
+      1,
+      0,
+      2,
+      0,
+      "IMAGE"
+    ],
+    [
+      3,
+      4,
+      0,
+      1,
+      0,
+      "CONDITIONING"
+    ],
+    [
+      4,
+      5,
+      0,
+      1,
+      1,
+      "CONDITIONING"
+    ]
+  ],
+  "groups": [],
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 1.0152559799477068,
+      "offset": [
+        58.35814123566077,
+        -65.12749162217047
+      ]
+    }
+  },
+  "version": 0.4
+}

asset/workflow/sdxl_base_mantra_tuner.json

@@ -0,0 +1,242 @@
+{
+  "last_node_id": 6,
+  "last_link_id": 5,
+  "nodes": [
+    {
+      "id": 2,
+      "type": "PreviewImage",
+      "pos": {
+        "0": 937,
+        "1": 243
+      },
+      "size": {
+        "0": 284.03680419921875,
+        "1": 246
+      },
+      "flags": {
+        "collapsed": false
+      },
+      "order": 4,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 1
+        }
+      ],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "PreviewImage"
+      },
+      "widgets_values": []
+    },
+    {
+      "id": 4,
+      "type": "ParameterNode",
+      "pos": {
+        "0": 0,
+        "1": 244
+      },
+      "size": {
+        "0": 311.9849853515625,
+        "1": 236.4765625
+      },
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            3
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ParameterNode"
+      },
+      "widgets_values": [
+        "ddim",
+        50,
+        5,
+        0.5,
+        "trailing",
+        1024,
+        1024,
+        2024
+      ]
+    },
+    {
+      "id": 5,
+      "type": "MantrasNode",
+      "pos": {
+        "0": 3,
+        "1": 547
+      },
+      "size": {
+        "0": 315,
+        "1": 58
+      },
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            4
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "MantrasNode"
+      },
+      "widgets_values": [
+        "Flat 2D Art"
+      ]
+    },
+    {
+      "id": 1,
+      "type": "ModelNode",
+      "pos": {
+        "0": 430,
+        "1": 243
+      },
+      "size": {
+        "0": 400,
+        "1": 234
+      },
+      "flags": {},
+      "order": 3,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "parameters",
+          "type": "CONDITIONING",
+          "link": 3,
+          "shape": 7
+        },
+        {
+          "name": "mantras",
+          "type": "CONDITIONING",
+          "link": 4,
+          "shape": 7
+        },
+        {
+          "name": "tuners",
+          "type": "CONDITIONING",
+          "link": 5,
+          "shape": 7
+        },
+        {
+          "name": "controls",
+          "type": "CONDITIONING",
+          "link": null,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            1
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ModelNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0",
+        "ModelScope",
+        "a cat",
+        ""
+      ]
+    },
+    {
+      "id": 6,
+      "type": "TunerNode",
+      "pos": {
+        "0": 8,
+        "1": 680
+      },
+      "size": {
+        "0": 315,
+        "1": 82
+      },
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            5
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "TunerNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0_Pencil Sketch Drawing",
+        1
+      ]
+    }
+  ],
+  "links": [
+    [
+      1,
+      1,
+      0,
+      2,
+      0,
+      "IMAGE"
+    ],
+    [
+      3,
+      4,
+      0,
+      1,
+      0,
+      "CONDITIONING"
+    ],
+    [
+      4,
+      5,
+      0,
+      1,
+      1,
+      "CONDITIONING"
+    ],
+    [
+      5,
+      6,
+      0,
+      1,
+      2,
+      "CONDITIONING"
+    ]
+  ],
+  "groups": [],
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 0.6934334949441366,
+      "offset": [
+        450.0923948318967,
+        30.783653239467935
+      ]
+    }
+  },
+  "version": 0.4
+}

asset/workflow/sdxl_base_mantra_tuner_control.json

@@ -0,0 +1,366 @@
+{
+  "last_node_id": 11,
+  "last_link_id": 9,
+  "nodes": [
+    {
+      "id": 1,
+      "type": "ModelNode",
+      "pos": {
+        "0": 439,
+        "1": 123
+      },
+      "size": {
+        "0": 400,
+        "1": 234
+      },
+      "flags": {},
+      "order": 6,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "parameters",
+          "type": "CONDITIONING",
+          "link": 3,
+          "shape": 7
+        },
+        {
+          "name": "mantras",
+          "type": "CONDITIONING",
+          "link": 4,
+          "shape": 7
+        },
+        {
+          "name": "tuners",
+          "type": "CONDITIONING",
+          "link": 5,
+          "shape": 7
+        },
+        {
+          "name": "controls",
+          "type": "CONDITIONING",
+          "link": 6,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            1
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ModelNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0",
+        "ModelScope",
+        "a cat",
+        ""
+      ]
+    },
+    {
+      "id": 5,
+      "type": "MantrasNode",
+      "pos": {
+        "0": 13,
+        "1": 336
+      },
+      "size": {
+        "0": 315,
+        "1": 58
+      },
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            4
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "MantrasNode"
+      },
+      "widgets_values": [
+        "Flat 2D Art"
+      ]
+    },
+    {
+      "id": 6,
+      "type": "TunerNode",
+      "pos": {
+        "0": 12,
+        "1": 455
+      },
+      "size": {
+        "0": 315,
+        "1": 82
+      },
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            5
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "TunerNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0_Pencil Sketch Drawing",
+        1
+      ]
+    },
+    {
+      "id": 11,
+      "type": "NoteNode",
+      "pos": {
+        "0": 845,
+        "1": 506
+      },
+      "size": {
+        "0": 398.3059387207031,
+        "1": 210.83267211914062
+      },
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "NoteNode"
+      },
+      "widgets_values": [
+        "This is a sample for quickly setting up ComfyUI using the Scepter open-source library:\n\n1) The first run involves downloading the models. By default, we will automatically pull models from ModelScope. The initial download may take some time, and you can also adjust the model source to change the model address.\n\n2) Currently, it supports various base models, basic settings for some inference hyperparameters, mantra settings, tuning model settings, and conditional generation node.\n\n3) In the future, we will gradually integrate interesting features to enrich the use cases.\n\n"
+      ]
+    },
+    {
+      "id": 8,
+      "type": "LoadImage",
+      "pos": {
+        "0": 471,
+        "1": 451
+      },
+      "size": {
+        "0": 315,
+        "1": 314
+      },
+      "flags": {},
+      "order": 3,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            7
+          ]
+        },
+        {
+          "name": "MASK",
+          "type": "MASK",
+          "links": null
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "LoadImage"
+      },
+      "widgets_values": [
+        "cat.jpg",
+        "image"
+      ]
+    },
+    {
+      "id": 7,
+      "type": "ControlNode",
+      "pos": {
+        "0": 15,
+        "1": 600
+      },
+      "size": {
+        "0": 330,
+        "1": 198
+      },
+      "flags": {},
+      "order": 5,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "source_image",
+          "type": "IMAGE",
+          "link": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            6
+          ]
+        },
+        {
+          "name": "Control Image",
+          "type": "IMAGE",
+          "links": [],
+          "slot_index": 1
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ControlNode"
+      },
+      "widgets_values": [
+        "SD_XL1.0_color",
+        "Color",
+        "CenterCrop",
+        1,
+        1024,
+        1024
+      ]
+    },
+    {
+      "id": 2,
+      "type": "PreviewImage",
+      "pos": {
+        "0": 959,
+        "1": 121
+      },
+      "size": {
+        "0": 284.03680419921875,
+        "1": 246
+      },
+      "flags": {
+        "collapsed": false
+      },
+      "order": 7,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 1
+        }
+      ],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "PreviewImage"
+      },
+      "widgets_values": []
+    },
+    {
+      "id": 4,
+      "type": "ParameterNode",
+      "pos": {
+        "0": 13,
+        "1": 43
+      },
+      "size": {
+        "0": 311.9849853515625,
+        "1": 236.4765625
+      },
+      "flags": {},
+      "order": 4,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "Result",
+          "type": "CONDITIONING",
+          "links": [
+            3
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ParameterNode"
+      },
+      "widgets_values": [
+        "ddim",
+        50,
+        5,
+        0.5,
+        "trailing",
+        1024,
+        1024,
+        2024
+      ]
+    }
+  ],
+  "links": [
+    [
+      1,
+      1,
+      0,
+      2,
+      0,
+      "IMAGE"
+    ],
+    [
+      3,
+      4,
+      0,
+      1,
+      0,
+      "CONDITIONING"
+    ],
+    [
+      4,
+      5,
+      0,
+      1,
+      1,
+      "CONDITIONING"
+    ],
+    [
+      5,
+      6,
+      0,
+      1,
+      2,
+      "CONDITIONING"
+    ],
+    [
+      6,
+      7,
+      0,
+      1,
+      3,
+      "CONDITIONING"
+    ],
+    [
+      7,
+      8,
+      0,
+      7,
+      0,
+      "IMAGE"
+    ]
+  ],
+  "groups": [],
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 0.7627768444385667,
+      "offset": [
+        254.62860058494985,
+        87.05734194144193
+      ]
+    }
+  },
+  "version": 0.4
+}

docs/en/tasks/ace.md

@@ -0,0 +1,184 @@
+<p align="center">
+
+  <h2 align="center"><img src="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/figures/icon.png" height=16> : All-round Creator and Editor Following <br> Instructions via Diffusion Transformer</h2>
+
+  <p align="center">
+    <a href="https://arxiv.org/abs/2410.00086"><img src='https://img.shields.io/badge/arXiv-ACE-red' alt='Paper PDF'></a>
+    <a href='https://ali-vilab.github.io/ace-page'><img src='https://img.shields.io/badge/Project_Page-ACE-blue' alt='Project Page'></a>
+    <a href='https://github.com/modelscope/scepter'><img src='https://img.shields.io/badge/Scepter-ACE-green'></a>
+    <a href='https://huggingface.co/spaces/scepter-studio/ACE-Chat'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Space-orange'></a>
+    <a href='https://huggingface.co/scepter-studio/ACE-0.6B-512px'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-orange'></a>
+    <a href='https://www.modelscope.cn/models/iic/ACE-0.6B-512px'><img src='https://img.shields.io/badge/ModelScope-Model-purple'></a>
+    <br>
+    <strong>Zhen Han*</strong>
+    ·
+    <strong>Zeyinzi Jiang*</strong>
+    ·
+    <strong>Yulin Pan*</strong>
+    ·
+    <strong>Jingfeng Zhang*</strong>
+    ·
+    <strong>Chaojie Mao*</strong>
+    <br>
+    <strong>Chenwei Xie</strong>
+    ·
+    <strong>Yu Liu</strong>
+    ·
+    <strong>Jingren Zhou</strong>
+    <br>
+    Tongyi Lab, Alibaba Group
+  </p>
+  <table align="center">
+    <tr>
+    <td>
+      <img src="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/figures/teaser.png">
+    </td>
+    </tr>
+  </table>
+
+
+## 🚀 Installation
+Install the necessary packages with `pip`: 
+```bash
+pip install -r requirements.txt
+```
+
+##  🔥 ACE Models
+|    **Model**     |                                                                                                                                                                                                            **Status**                                                                                                                                                                                                             | 
+|:----------------:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
+|  ACE-0.6B-512px  |          [![Demo link](https://img.shields.io/badge/Demo-ACE_Chat-purple)](https://huggingface.co/spaces/scepter-studio/ACE-Chat)<br>[![ModelScope link](https://img.shields.io/badge/ModelScope-Model-blue)](https://www.modelscope.cn/models/iic/ACE-0.6B-512px)  [![HuggingFace link](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-yellow)](https://huggingface.co/scepter-studio/ACE-0.6B-512px)          |
+| ACE-0.6B-1024px  | [![Demo link](https://img.shields.io/badge/Demo-ACE_Refiner_Chat-purple)](https://huggingface.co/spaces/scepter-studio/ACE-Refiner-Chat)<br>[![ModelScope link](https://img.shields.io/badge/ModelScope-Model-blue)](https://www.modelscope.cn/models/iic/ACE-0.6B-1024px)  [![HuggingFace link](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-yellow)](https://huggingface.co/scepter-studio/ACE-0.6B-1024px) |             |
+## 🖼 Model Performance Visualization
+
+The current model's parameters scale of ACE is 0.6B, which imposes certain limitations on the quality of image generation. [FLUX.1-Dev](https://huggingface.co/black-forest-labs/FLUX.1-dev), on the other hand, 
+has a significant advantage in text-to-image generation quality. By using SDEdit, we can effectively leverage the generative capabilities of FLUX to further enhance the image results generated by ACE. Based on the above considerations, we have designed the ACE-Refiner pipeline, as shown in the diagram below.
+
+![ACE_REFINER](https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/ace_method/ace_refiner_process.webp)
+
+As shown in the figure below, when the strength 
+σ of the generated image is high, the generated image will suffer from fidelity loss compared to the original image. Conversely, lower 
+σ does not significantly improve the image quality. Therefore, users can make a trade-off between fidelity to the generated result and the image quality based on their own needs. 
+Users can set the value of "REFINER_SCALE" in the configuration file `config/inference_config/models/ace_0.6b_1024_refiner.yaml`. 
+We recommend that users use the advance options in the [webui-demo](#-chat-bot-) for effect verification.
+
+![ACE_REFINER_EXAMPLE](https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/ace_method/ace_refiner.webp)
+
+
+We compared the generation and editing performance of different models on several tasks, as shown as following. 
+![Samples](https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/ace_method/samples_compare.webp)
+
+
+## 🔥 Training
+
+We offer a demonstration training YAML that enables the end-to-end training of ACE using a toy dataset. For a comprehensive overview of the hyperparameter configurations, please consult `config/ace_0.6b_512_train.yaml`.
+
+### Prepare datasets
+
+Please find the dataset class located in `modules/data/dataset/dataset.py`, 
+designed to facilitate end-to-end training using an open-source toy dataset. 
+Download a dataset zip file from [modelscope](https://www.modelscope.cn/models/iic/scepter/resolve/master/datasets/hed_pair.zip), and then extract its contents into the `cache/datasets/` directory.
+
+Should you wish to prepare your own datasets, we recommend consulting `modules/data/dataset/dataset.py` for detailed guidance on the required data format.
+
+### Prepare initial weight
+The ACE checkpoint has been uploaded to both ModelScope and HuggingFace platforms:
+* [ModelScope](https://www.modelscope.cn/models/iic/ACE-0.6B-512px)
+* [HuggingFace](https://huggingface.co/scepter-studio/ACE-0.6B-512px)
+
+In the provided training YAML configuration, we have designated the Modelscope URL as the default checkpoint URL. Should you wish to transition to Hugging Face, you can effortlessly achieve this by modifying the PRETRAINED_MODEL value within the YAML file (replace the prefix "ms://iic" to "hf://scepter-studio").
+
+
+### Start training
+
+You can easily start training procedure by executing the following command:
+```bash
+# ACE-0.6B-512px
+PYTHONPATH=. python tools/run_train.py --cfg config/ace_0.6b_512_train.yaml
+# ACE-0.6B-1024px
+PYTHONPATH=. python tools/run_train.py --cfg config/ace_0.6b_1024_train.yaml
+```
+
+## 🚀 Inference
+
+We provide a simple inference demo that allows users to generate images from text descriptions.
+```bash
+ PYTHONPATH=. python tools/run_inference.py --cfg config/inference_config/models/ace_0.6b_512.yaml --instruction "make the boy cry, his eyes filled with tears" --seed 199999 --input_image examples/input_images/example0.webp
+```
+We recommend runing the examples for quick testing. Running the following command will run the example inference and the results will be saved in `examples/output_images/`.
+```bash
+PYTHONPATH=. python tools/run_inference.py --cfg config/inference_config/models/ace_0.6b_512.yaml
+```
+
+## 💬 Chat Bot 
+We have developed an chatbot UI utilizing Gradio, designed to transform user input in natural language into visually stunning images that align semantically with the provided instructions. Users can effortlessly initiate the chatbot app by executing the following command:
+```bash
+python chatbot/run_gradio.py --cfg chatbot/config/chatbot_ui.yaml --server_port 2024
+```
+
+<table align="center">
+  <tr>
+  <td>
+    <img src="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/videos/demo_chat.gif">
+  </td>
+  </tr>
+</table>
+
+## ⚙️️ ComfyUI Workflow
+
+![Workflow](https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_example.jpg)
+
+We support the use of ACE in the ComfyUI Workflow through the following methods:
+
+1) Automatic installation directly via the ComfyUI Manager by searching for the **ComfyUI-Scepter** node.
+2) Manually install by moving custom_nodes from Scepter to ComfyUI.
+```shell
+git clone https://github.com/modelscope/scepter.git
+cd path/to/scepter
+pip install -e .
+cp -r path/to/scepter/workflow/ path/to/ComfyUI/custom_nodes/ComfyUI-Scepter
+cd path/to/ComfyUI
+python main.py
+```
+
+**Note**: You can use the nodes by dragging the sample images below into ComfyUI. Additionally, our nodes can automatically pull models from ModelScope or HuggingFace by selecting the *model_source* field, or you can place the already downloaded models in a local path.
+
+<table><tbody>
+  <tr>
+    <th align="center" colspan="4">ACE Workflow Examples</th>
+  </tr>
+  <tr>
+    <th align="center" colspan="1">Control</th>
+    <th align="center" colspan="1">Semantic</th>
+    <th align="center" colspan="1">Element</th>
+  </tr>
+  <tr>
+    <td>
+      <a href="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_control.png" target="_blank">
+        <img src="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_control.png" width="200">
+      </a>
+    </td>
+    <td>
+      <a href="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_semantic.png" target="_blank">
+        <img src="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_semantic.png" width="200">
+      </a>
+    </td>
+    <td>
+      <a href="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_element.png" target="_blank">
+        <img src="https://raw.githubusercontent.com/ali-vilab/ACE/refs/heads/main/assets/comfyui/ace_element.png" width="200">
+      </a>
+    </td>
+  </tr>
+</tbody>
+</table>
+
+
+## 📝 Citation
+
+```bibtex
+@article{han2024ace,
+  title={ACE: All-round Creator and Editor Following Instructions via Diffusion Transformer},
+  author={Han, Zhen and Jiang, Zeyinzi and Pan, Yulin and Zhang, Jingfeng and Mao, Chaojie and Xie, Chenwei and Liu, Yu and Zhou, Jingren},
+  journal={arXiv preprint arXiv:2410.00086},
+  year={2024}
+}
+```

docs/en/tasks/stylebooth.md

@@ -56,7 +56,7 @@ Given an original image, image editing aims to generate an image that align with
 - More models will be released in the future.
 
 ## Run StyleBooth
-- Code implementation: See model configuration and code based on [🪄SCEPTER](https://github.com/modelscope/scepter).
+- Code implementation: See model configuration and code based on [🪄SCEPTER](https://github.com/modelscope/scepter/blob/main/docs/en/tasks/stylebooth.md).
 
 - Demo: Try [🖥️SCEPTER Studio](https://github.com/modelscope/scepter/tree/main?tab=readme-ov-file#%EF%B8%8F-scepter-studio).
 
@@ -114,3 +114,17 @@ class DiffusionInferenceTest(unittest.TestCase):
 if __name__ == '__main__':
     unittest.main()
 ```
+
+## StyleTuner and De-StyleTuner.
+
+### Base I2I Model.
+
+For style and de-style tuning, we use a private high-resolution I2I model trained with [InstructPix2Pix dataset](https://instruct-pix2pix.eecs.berkeley.edu/) as base model. However, one can try the same tunning process using this [yaml](https://github.com/modelscope/scepter/blob/main/scepter/methods/edit/edit_512_lora.yaml) based on any other I2I model, such as StyleBooth (shown in this yaml), [InstructPix2Pix](https://github.com/timothybrooks/instruct-pix2pix) or [MagicBrush](https://github.com/OSU-NLP-Group/MagicBrush).
+
+### Training Data.
+
+Please check the zips for correct format: [De-Text](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fdetext.zip), [Image2Hed](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fhed_pair.zip), [Image2Depth](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fimage2depth.zip), [Depth2Image](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fdepth2image.zip).
+
+### Launch.
+
+See [code](https://github.com/modelscope/scepter/blob/cbbef4a2da5b66fc33b9f8ece7f2fb4aac9d6e3c/tests/tools/test_train.py#L220) for more information.

environment.yaml

@@ -1,10 +0,0 @@
-name: scepter
-channels:
-  - defaults
-dependencies:
-  - python==3.8
-  - pip>=20.3
-  - numpy>=1.23.1
-  - pip:
-      - -r requirements/recommended.txt
-      - -r requirements.txt

install.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) Alibaba, Inc. and its affiliates.
+import os
+import shutil
+import subprocess
+import sys
+
+if sys.argv[0] == 'install.py':
+    sys.path.append('.')   # for portable version
+
+source_folder = os.path.join(os.path.dirname(__file__), "scepter/workflow")
+current_dir = os.path.dirname(__file__)
+destination_folder = os.path.join(os.path.dirname(current_dir), "ComfyUI-Scepter")
+
+if not os.path.exists(destination_folder):
+    shutil.copytree(source_folder, destination_folder)
+    print(f"{os.path.abspath(source_folder)} copy to {os.path.abspath(destination_folder)} success!")
+else:
+    print(f"{os.path.abspath(destination_folder)} exist.")
+
+# pip install scepter
+subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'scepter'])

readme.md

@@ -14,10 +14,16 @@ SCEPTER integrates popular community-driven implementations as well as proprieta
 SCEPTER offers 3 core components:
 - [Generative training and inference framework](#tutorials)
 - [Easy implementation of popular approaches](#currently-supported-approaches)
-- [Interactive user interface: SCEPTER Studio](#launch)
+- [Interactive user interface: SCEPTER Studio & Comfy UI](#launch)
 
 
 ## 🎉 News
+- [🔥🔥🔥 2025.01]: We report ACE++, an instruction-based diffusion framework that tackles various image generation and editing tasks. The code and paper is available on [ACE++](https://ali-vilab.github.io/ACE_plus_page/).
+- [2024.11]: Supports video files, video annotation, caption translation in data management, and inference & training of the [CogVideoX](https://arxiv.org/abs/2408.06072).
+- [2024.10]: We are pleased to announce the release of the code for [ACE](https://arxiv.org/abs/2410.00086), supporting Customized Training / Comfy UI Workflow / gradio-based ChatBot Interface. 
+- [2024.10]: Support for inference and tuning with [FLUX](https://huggingface.co/black-forest-labs/FLUX.1-dev), as well as for building [ComfyUI](https://github.com/comfyanonymous/ComfyUI) workflows using this framework.
+- [2024.09]: We introduce **ACE**, an **A**ll-round **C**reator and **E**ditor adept at executing a diverse array of image editing tasks tailored to your specifications. Built upon the cutting-edge Diffusion Transformer architecture, ACE has been extensively trained on a comprehensive dataset to seamlessly interpret and execute any natural language instruction. For further information, please consult the [project page](https://ali-vilab.github.io/ace-page/).
+- [2024.07]: Support the inference and training of open-source generative models based on the [DiT](https://arxiv.org/abs/2212.09748) architecture, such as [SD3](https://arxiv.org/pdf/2403.03206) and [PixArt](https://arxiv.org/abs/2310.00426).
 - [2024.05]: Introducing SCEPTER v1, supporting customized image edit tasks! Simply provide 10 image pairs, SCEPTER will tune an edit tuner for your own Image-to-Image tasks, like `Clay Style`, `De-Text`, `Segmentation`, etc.
 - [2024.04]: New [StyleBooth](https://ali-vilab.github.io/stylebooth-page/) demo on SCEPTER Studio for`Text-Based Style Editing`.
 - [2024.03]: We optimize the training UI and checkpoint management. New [LAR-Gen](https://arxiv.org/abs/2403.19534) model has been added on SCEPTER Studio, supporting `zoom-out`, `virtual try on`, `inpainting`.
@@ -27,58 +33,108 @@ SCEPTER offers 3 core components:
 - [2023.12]: We propose [SCEdit](https://arxiv.org/abs/2312.11392), an efficient and controllable generation framework.
 - [2023.12]: We release [🪄SCEPTER](https://github.com/modelscope/scepter/) library.
 
+[//]: # (## 🖼 Gallery for Recent Works)
 
-## 🖼 Gallery for Recent Works
+[//]: # ()
+[//]: # (### FLUX Tuners)
 
-### Edit Tuners
+[//]: # ()
+[//]: # (<table><tbody>)
 
-Simply provide 10 image pairs, SCEPTER will tune an edit tuner for your own Image-to-Image tasks, like `Clay Style`, `De-Text`, `Segmentation`, etc.
-Try our official few-shot datasets: [De-Text](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fdetext.zip), [Image2Hed](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fhed_pair.zip), [Image2Depth](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fimage2depth.zip), [Depth2Image](https://www.modelscope.cn/api/v1/models/iic/scepter/repo?Revision=master&FilePath=datasets%2Fdepth2image.zip).
+[//]: # (  <tr>)
 
+[//]: # (    <th align="center" colspan="3">Yarn Style</th>)
+
+[//]: # (    <th align="center" colspan="3">Soft Watercolor Style</th>)
+
+[//]: # (  </tr>)
+
+[//]: # (  <tr>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_2_1.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_2_2.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_2_3.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_1_1.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_1_2.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_1_3.webp" width="200"></td>)
+
+[//]: # (  </tr>)
+
+[//]: # (  <tr>)
+
+[//]: # (    <th align="center" colspan="3">Travel Style</th>)
+
+[//]: # (    <th align="center" colspan="3">WuKong Style</th>)
+
+[//]: # (  </tr>)
+
+[//]: # (  <tr>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_3_1.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_3_2.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_3_3.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_4_1.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_4_2.webp" width="200"></td>)
+
+[//]: # (    <td><img src="asset/images/flux_tuner/flux_tuner_4_3.webp" width="200"></td>)
+
+[//]: # (  </tr>)
+
+[//]: # (</tbody>)
+
+[//]: # (</table>)
+
+### ComfyUI Workflow
+
+![Workflow](asset/workflow/workflow.jpg)
 
 <table><tbody>
   <tr>
-    <th align="center" colspan="4">Clay Style<br>Prompt: "Convert this image into clay style"</th>
+    <th align="center" colspan="4">Example Workflow Case</th>
   </tr>
   <tr>
-    <td><img src="asset/images/edit_tuner/vermeer.jpeg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/clay_vermeer.jpeg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/cat_512.jpg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/clay_cat.jpeg" width="300"></td>
+    <th align="center" colspan="1">Base</th>
+    <th align="center" colspan="1">+Mantra</th>
+    <th align="center" colspan="1">+Tuner</th>
+    <th align="center" colspan="1">+Control</th>
   </tr>
   <tr>
-    <th align="center" colspan="2">De-Text<br>Prompt: "Remove the texts"</th>
-    <th align="center" colspan="2">Image2Hed<br>Prompt: "Convert to an edge map"</th>
-  </tr>
-  <tr>
-    <td><img src="asset/images/edit_tuner/text.jpg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/detext.jpeg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/cat_512.jpg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/hed.jpeg" width="300"></td>
-  </tr>
-    <tr>
-    <th align="center" colspan="2">Image2Depth<br>Prompt: "Calculate the depth map"</th>
-    <th align="center" colspan="2">Depth2Image<br>Prompt: "Convert depth map into color image"</th>
-  </tr>
-  <tr>
-    <td><img src="asset/images/edit_tuner/house.jpg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/image2depth.jpeg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/depth.jpg" width="300"></td>
-    <td><img src="asset/images/edit_tuner/depth2image.jpeg" width="300"></td>
+    <td>
+      <a href="asset/workflow/sdxl_base.json" target="_blank">
+        <img src="asset/workflow/sdxl_base.jpg" width="200">
+      </a>
+    </td>
+    <td>
+      <a href="asset/workflow/sdxl_base_mantra.json" target="_blank">
+        <img src="asset/workflow/sdxl_base_mantra.jpg" width="200">
+      </a>
+    </td>
+    <td>
+      <a href="asset/workflow/sdxl_base_mantra_tuner.json" target="_blank">
+        <img src="asset/workflow/sdxl_base_mantra_tuner.jpg" width="200">
+      </a>
+    </td>
+    <td>
+      <a href="asset/workflow/sdxl_base_mantra_tuner_control.json" target="_blank">
+        <img src="asset/workflow/sdxl_base_mantra_tuner_control.jpg" width="200">
+      </a>
+    </td>
   </tr>
 </tbody>
 </table>
-Note: Left image is input and right image is output.
+
 
 ## 🛠️ Installation
 
-- Create new environment with `conda` command:
-
-```shell
-conda env create -f environment.yaml
-conda activate scepter
-```
-
 - Install with `pip` command:
 
 We recommend installing the specific version of PyTorch and accelerate toolbox [xFormers](https://pypi.org/project/xformers/). You can install these recommended version by pip:
@@ -102,16 +158,19 @@ pip install scepter
 
 ### Currently supported approaches
 
-|            Tasks             |                   Methods                    | Links                                                                                                                                                                                                                                                       |
-|:----------------------------:|:--------------------------------------------:|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-|   Text-to-image generation   |                   SD v1.5                    | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/runwayml/stable-diffusion-v1-5)                                                                                                         |
-|   Text-to-image generation   |                   SD v2.1                    | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/runwayml/stable-diffusion-v1-5)                                                                                                         |
-|   Text-to-image generation   |                    SD-XL                     | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)                                                                                               |
-|       Efficient Tuning       |                     LoRA                     | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=LoRA&color=red&logo=arxiv)](https://arxiv.org/abs/2106.09685)                                                                                                                         |
-|       Efficient Tuning       |            Res-Tuning(NeurIPS23)             | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=Res-Tuing&color=red&logo=arxiv)](https://arxiv.org/abs/2310.19859) [![Page link](https://img.shields.io/badge/Page-ResTuning-Gree)](https://res-tuning.github.io/)                    |
-| Controllable image synthesis | [🌟SCEdit(CVPR24)](docs/en/tasks/scedit.md)  | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=SCEdit&color=red&logo=arxiv)](https://arxiv.org/abs/2312.11392)   [![Page link](https://img.shields.io/badge/Page-SCEdit-Gree)](https://scedit.github.io/)                            |
-|        Image editing         |     [🌟LAR-Gen](docs/en/tasks/largen.md)     | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=LARGen&color=red&logo=arxiv)](https://arxiv.org/abs/2403.19534)   [![Page link](https://img.shields.io/badge/Page-LARGen-Gree)](https://ali-vilab.github.io/largen-page/)             |
-|        Image editing         |  [🌟StyleBooth](docs/en/tasks/stylebooth.md) | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=StyleBooth&color=red&logo=arxiv)](https://arxiv.org/abs/2404.12154)   [![Page link](https://img.shields.io/badge/Page-StyleBooth-Gree)](https://ali-vilab.github.io/stylebooth-page/) |
+|            Tasks             |                     Methods                      | Links                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             |
+|:----------------------------:|:------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+|   Text-to-image Generation   |                     SD v1.5                      | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/runwayml/stable-diffusion-v1-5)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
+|   Text-to-image Generation   |                     SD v2.1                      | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/runwayml/stable-diffusion-v1-5)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
+|   Text-to-image Generation   |                      SD-XL                       | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     |
+|   Text-to-image Generation   |                       FLUX                       | [![Hugging Face Repo](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Repo-blue)](https://huggingface.co/black-forest-labs/FLUX.1-dev)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
+|       Efficient Tuning       |                       LoRA                       | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=LoRA&color=red&logo=arxiv)](https://arxiv.org/abs/2106.09685)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
+|       Efficient Tuning       |              Res-Tuning(NeurIPS23)               | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=Res-Tuing&color=red&logo=arxiv)](https://arxiv.org/abs/2310.19859) [![Page link](https://img.shields.io/badge/Page-ResTuning-Gree)](https://res-tuning.github.io/)                                                                                                                                                                                                                                                                                                                                                                                                                          |
+| Controllable Image Synthesis |   [🌟SCEdit(CVPR24)](docs/en/tasks/scedit.md)    | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=SCEdit&color=red&logo=arxiv)](https://arxiv.org/abs/2312.11392)   [![Page link](https://img.shields.io/badge/Page-SCEdit-Gree)](https://scedit.github.io/)                                                                                                                                                                                                                                                                                                                                                                                                                                  |
+|        Image Editing         |       [🌟LAR-Gen](docs/en/tasks/largen.md)       | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=LARGen&color=red&logo=arxiv)](https://arxiv.org/abs/2403.19534)   [![Page link](https://img.shields.io/badge/Page-LARGen-Gree)](https://ali-vilab.github.io/largen-page/)                                                                                                                                                                                                                                                                                                                                                                                                                   |
+|        Image Editing         |   [🌟StyleBooth](docs/en/tasks/stylebooth.md)    | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=StyleBooth&color=red&logo=arxiv)](https://arxiv.org/abs/2404.12154)   [![Page link](https://img.shields.io/badge/Page-StyleBooth-Gree)](https://ali-vilab.github.io/stylebooth-page/)                                                                                                                                                                                                                                                                                                                                                                                                       |
+| Image Generation and Editing |  [🌟ACE](https://ali-vilab.github.io/ace-page/)  | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=ACE&color=red&logo=arxiv)](https://arxiv.org/abs/2410.00086)   [![Page link](https://img.shields.io/badge/Page-ACE-Gree)](https://ali-vilab.github.io/ace-page/) [![Demo link](https://img.shields.io/badge/Demo-ACE-purple)](https://huggingface.co/spaces/scepter-studio/ACE-Chat) <br> [![ModelScope link](https://img.shields.io/badge/ModelScope-Model-blue)](https://www.modelscope.cn/models/iic/ACE-0.6B-512px)  [![HuggingFace link](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-yellow)](https://huggingface.co/scepter-studio/ACE-0.6B-512px)               |
+| Image Generation and Editing | [🌟ACE++](https://ali-vilab.github.io/ACE_plus_page/) | [![Arxiv   link](https://img.shields.io/static/v1?label=arXiv&message=ACEPlus&color=red&logo=arxiv)](https://arxiv.org/abs/2501.02487)   [![Page link](https://img.shields.io/badge/Page-ACE++-Gree)](https://ali-vilab.github.io/ACE_plus_page/) [![Demo link](https://img.shields.io/badge/Demo-ACE++-purple)](https://huggingface.co/spaces/scepter-studio/ACE-Plus) <br> [![ModelScope link](https://img.shields.io/badge/ModelScope-Model-blue)](https://www.modelscope.cn/models/iic/ACE_Plus/summary)  [![HuggingFace link](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-yellow)](https://huggingface.co/ali-vilab/ACE_Plus/tree/main) |
 
 
 ## 🖥️ SCEPTER Studio
@@ -145,6 +204,24 @@ Therefore, subsequent startups will become much faster (about one minute) as dow
 We deploy a work studio on Modelscope that includes only the inference tab, please refer to [ms_scepter_studio](https://www.modelscope.cn/studios/iic/scepter_studio/summary) and [hf_scepter_studio](https://huggingface.co/spaces/modelscope/scepter_studio)
 
 
+
+## ⚙️️ ComfyUI Workflow
+
+We support the use of all models in the ComfyUI Workflow through the following methods:
+
+1) Automatic installation directly via the ComfyUI Manager by searching for the **ComfyUI-Scepter** node.
+2) Manually install by moving custom_nodes from Scepter to ComfyUI.
+```shell
+git clone https://github.com/modelscope/scepter.git
+cd path/to/scepter
+pip install -e .
+cp -r path/to/scepter/workflow/ path/to/ComfyUI/custom_nodes/ComfyUI-Scepter
+cd path/to/ComfyUI
+python main.py
+```
+
+**Note**: You can use the nodes by dragging the sample images into ComfyUI. Additionally, our nodes can automatically pull models from ModelScope or HuggingFace by selecting the *model_source* field, or you can place the already downloaded models in a local path.
+
 ## 🔍 Learn More
 
 - [Alibaba TongYi Vision Intelligence Lab](https://github.com/ali-vilab)
@@ -177,4 +254,4 @@ This project is licensed under the [Apache License (Version 2.0)](https://github
 
 
 ## Acknowledgement
-Thanks to [Stability-AI](https://github.com/Stability-AI), [SWIFT library](https://github.com/modelscope/swift/) and [Fooocus](https://github.com/lllyasviel/Fooocus) for their awesome work.
+Thanks to [Stability-AI](https://github.com/Stability-AI), [SWIFT library](https://github.com/modelscope/swift/), [Fooocus](https://github.com/lllyasviel/Fooocus) and [ComfyUI](https://github.com/comfyanonymous/ComfyUI) for their awesome work.

requirements/framework.txt

@@ -1,8 +1,8 @@
 albumentations
-bezier
+beautifulsoup4
 einops
-modelscope
-ms-swift>=2.0.1
+modelscope[framework]
+ms-swift
 numpy
 open_clip_torch
 opencv-python
@@ -11,5 +11,7 @@ oss2>=2.15.0
 pycocotools
 pyyaml>=5.3.1
 scikit-image
+scikit-learn
+sentencepiece
 torchsde
 transformers

requirements/recommended.txt

@@ -1,4 +1,5 @@
 git+https://github.com/cocodataset/panopticapi.git
-torch==2.0.1
-torchvision==0.15.2
-xformers==0.0.21
+torch==2.4.1
+torchvision==0.19.1
+flash-attn==2.5.8
+xformers==0.0.28

requirements/scepter_studio.txt

@@ -1,5 +1,6 @@
 bitsandbytes
-gradio>=3.47.1,<4.0.0
+gradio
+gradio_imageslider
 imagehash
 psutil
 tiktoken

scepter/__init__.py

@@ -1,18 +1,28 @@
 # -*- coding: utf-8 -*-
 # Copyright (c) Alibaba, Inc. and its affiliates.
-import os
+from typing import TYPE_CHECKING
+from scepter.modules.utils.import_utils import LazyImportModule
 
-import scepter
-from scepter.modules import data, model, opt, solver, transform, utils
-from scepter.tools.helper import get_module_list as module_list
-from scepter.tools.helper import \
-    get_module_object_config as configures_by_objects
-from scepter.tools.helper import get_module_objects as objects_by_module
-from scepter.version import __version__, version_info
 
-dirname = os.path.dirname(scepter.__file__)
+if TYPE_CHECKING:
+    from scepter.modules import data, model, opt, solver, transform, utils
+    from scepter.tools.helper import get_module_list as module_list
+    from scepter.tools.helper import \
+        get_module_object_config as configures_by_objects
+    from scepter.tools.helper import get_module_objects as objects_by_module
+    from scepter.version import __version__, version_info
+else:
+    _import_structure = {
+        'modules': ['data', 'model', 'opt', 'solver', 'transform', 'utils'],
+        'helper': ['get_module_list', 'get_module_object_config', 'get_module_objects'],
+        'version': ['__version__', 'version_info']
+    }
 
-__all__ = [
-    utils, transform, data, model, solver, version_info, opt, '__version__',
-    'dirname'
-]
+    import sys
+    sys.modules[__name__] = LazyImportModule(
+        __name__,
+        globals()['__file__'],
+        _import_structure,
+        module_spec=__spec__,
+        extra_objects={},
+    )

scepter/methods/edit/dit_ace_0.6b_1024.yaml

@@ -0,0 +1,161 @@
+ENV:
+  BACKEND: nccl
+  SEED: 2024
+#
+SOLVER:
+  NAME: ACESolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 500
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: 50
+  RESCALE_LR: False
+  #
+  WORK_DIR: ./cache/save_data/ace_0.6b_1024
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    - NAME: "HuggingfaceFs"
+      TEMP_DIR: ./cache/cache_data
+    - NAME: "LocalFs"
+      TEMP_DIR: ./cache/cache_data
+    - NAME: "ModelscopeFs"
+      TEMP_DIR: ./cache/cache_data
+
+  #
+  MODEL:
+    NAME: LatentDiffusionACE
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 0.18215
+    SIZE_FACTOR: 8
+    DECODER_BIAS: 0.5
+    DEFAULT_N_PROMPT:
+    USE_EMA: True
+    EVAL_EMA: False
+    TEXT_IDENTIFIER: [ '{image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+    USE_TEXT_POS_EMBEDDINGS: True
+    #
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: eps
+      MIN_SNR_GAMMA:
+      NOISE_SCHEDULER:
+        NAME: LinearScheduler
+        NUM_TIMESTEPS: 1000
+        BETA_MIN: 0.0001
+        BETA_MAX: 0.02
+    #
+    DIFFUSION_MODEL:
+      NAME: ACE
+      PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/dit/ace_0.6b_1024px.pth
+      IGNORE_KEYS: [ ]
+      PATCH_SIZE: 2
+      IN_CHANNELS: 4
+      HIDDEN_SIZE: 1152
+      DEPTH: 28
+      NUM_HEADS: 16
+      MLP_RATIO: 4.0
+      PRED_SIGMA: True
+      DROP_PATH: 0.0
+      WINDOW_DIZE: 0
+      Y_CHANNELS: 4096
+      MAX_SEQ_LEN: 4096
+      QK_NORM: True
+      USE_GRAD_CHECKPOINT: True
+      ATTENTION_BACKEND: flash_attn
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      EMBED_DIM: 4
+      PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/vae/vae.bin
+      IGNORE_KEYS: []
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/text_encoder/t5-v1_1-xxl/
+      TOKENIZER_PATH: ms://iic/ACE-0.6B-1024px@models/tokenizer/t5-v1_1-xxl
+      LENGTH: 120
+      T5_DTYPE: bfloat16
+      ADDED_IDENTIFIER: [ '{image}', '{caption}', '{mask}', '{ref_image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+      CLEAN: whitespace
+      USE_GRAD: False
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 20
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 1e-7
+    EPS: 1e-10
+    WEIGHT_DECAY: 5e-4
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDatasetForACE
+    MODE: train
+    MS_DATASET_NAME: cache/datasets/hed_pair
+    MS_DATASET_NAMESPACE: ""
+    MS_DATASET_SPLIT: "train"
+    MS_DATASET_SUBNAME: ""
+    PROMPT_PREFIX: ""
+    REPLACE_STYLE: False
+    MAX_SEQ_LEN: 4096
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 1
+    SAMPLER:
+      NAME: LoopSampler
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 0
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 50
+    -
+      NAME: CheckpointHook
+      INTERVAL: 100
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 100

scepter/methods/edit/dit_ace_0.6b_512.yaml

@@ -0,0 +1,161 @@
+ENV:
+  BACKEND: nccl
+  SEED: 2024
+#
+SOLVER:
+  NAME: ACESolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 500
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: 50
+  RESCALE_LR: False
+  #
+  WORK_DIR: ./cache/save_data/ace_0.6b_512
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    - NAME: "HuggingfaceFs"
+      TEMP_DIR: ./cache/cache_data
+    - NAME: "LocalFs"
+      TEMP_DIR: ./cache/cache_data
+    - NAME: "ModelscopeFs"
+      TEMP_DIR: ./cache/cache_data
+
+  #
+  MODEL:
+    NAME: LatentDiffusionACE
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 0.18215
+    SIZE_FACTOR: 8
+    DECODER_BIAS: 0.5
+    DEFAULT_N_PROMPT:
+    USE_EMA: True
+    EVAL_EMA: False
+    TEXT_IDENTIFIER: [ '{image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+    USE_TEXT_POS_EMBEDDINGS: True
+    #
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: eps
+      MIN_SNR_GAMMA:
+      NOISE_SCHEDULER:
+        NAME: LinearScheduler
+        NUM_TIMESTEPS: 1000
+        BETA_MIN: 0.0001
+        BETA_MAX: 0.02
+    #
+    DIFFUSION_MODEL:
+      NAME: ACE
+      PRETRAINED_MODEL: ms://iic/ACE-0.6B-512px@models/dit/ace_0.6b_512px.pth
+      IGNORE_KEYS: [ ]
+      PATCH_SIZE: 2
+      IN_CHANNELS: 4
+      HIDDEN_SIZE: 1152
+      DEPTH: 28
+      NUM_HEADS: 16
+      MLP_RATIO: 4.0
+      PRED_SIGMA: True
+      DROP_PATH: 0.0
+      WINDOW_DIZE: 0
+      Y_CHANNELS: 4096
+      MAX_SEQ_LEN: 1024
+      QK_NORM: True
+      USE_GRAD_CHECKPOINT: True
+      ATTENTION_BACKEND: flash_attn
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      EMBED_DIM: 4
+      PRETRAINED_MODEL: ms://iic/ACE-0.6B-512px@models/vae/vae.bin
+      IGNORE_KEYS: []
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://iic/ACE-0.6B-512px@models/text_encoder/t5-v1_1-xxl/
+      TOKENIZER_PATH: ms://iic/ACE-0.6B-512px@models/tokenizer/t5-v1_1-xxl
+      LENGTH: 120
+      T5_DTYPE: bfloat16
+      ADDED_IDENTIFIER: [ '{image}', '{caption}', '{mask}', '{ref_image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+      CLEAN: whitespace
+      USE_GRAD: False
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 20
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 1e-7
+    EPS: 1e-10
+    WEIGHT_DECAY: 5e-4
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDatasetForACE
+    MODE: train
+    MS_DATASET_NAME: cache/datasets/hed_pair
+    MS_DATASET_NAMESPACE: ""
+    MS_DATASET_SPLIT: "train"
+    MS_DATASET_SUBNAME: ""
+    PROMPT_PREFIX: ""
+    REPLACE_STYLE: False
+    MAX_SEQ_LEN: 1024
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 1
+    SAMPLER:
+      NAME: LoopSampler
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 0
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 50
+    -
+      NAME: CheckpointHook
+      INTERVAL: 100
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 100

scepter/methods/edit/edit_512_lora.yaml

@@ -20,7 +20,7 @@ SOLVER:
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -

scepter/methods/examples/classification/example.yaml

@@ -11,7 +11,7 @@ SOLVER:
   # NUM_FOLDS DESCRIPTION: Num folds for training. TYPE: int default: 0
   NUM_FOLDS: 1
   # WORK_DIR DESCRIPTION: Save dir of the training log or model. TYPE: str default: ''
-  WORK_DIR: ./exp12/
+  WORK_DIR: ./cache/save_data/example/
   LOG_FILE: std_log.txt
   # EVAL_INTERVAL DESCRIPTION: Eval the model interval. TYPE: int default: 1
   EVAL_INTERVAL: 1
@@ -102,7 +102,7 @@ SOLVER:
     # DATASET DESCRIPTION: the public dataset name TYPE: str default: 'cifar10'
     DATASET: cifar10
     # DATA_ROOT DESCRIPTION: the download data save path TYPE: str default: ''
-    DATA_ROOT: ./local_data/cifar10
+    DATA_ROOT: ./cache/cache_data/cifar10
     # MODE DESCRIPTION: test TYPE: str default: test
     MODE: test
     # PIN_MEMORY DESCRIPTION: pin_memory for data loader TYPE: bool default: False

scepter/methods/examples/generation/dit_cogvideox1.5_5b_i2v_lora.yaml

@@ -0,0 +1,277 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox1.5_5b_i2v_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  FPS: 16
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 64
+      LORA_ALPHA: 64
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 0.7  
+    NOISED_IMAGE_DROPOUT: 0.05
+    INVERT_SCALE_LATENTS: True
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      USE_DYNAMIC_CFG: False
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 1.0  
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: # 5b-I2V diff
+        - ms://ZhipuAI/CogVideoX1.5-5B-I2V@transformer/diffusion_pytorch_model-00001-of-00003.safetensors
+        - ms://ZhipuAI/CogVideoX1.5-5B-I2V@transformer/diffusion_pytorch_model-00002-of-00003.safetensors
+        - ms://ZhipuAI/CogVideoX1.5-5B-I2V@transformer/diffusion_pytorch_model-00003-of-00003.safetensors
+      NUM_ATTENTION_HEADS: 48  
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 32
+      LATENT_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      OFS_EMBED_DIM: 512 # v1.5 diff
+      NUM_LAYERS: 42  
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 300
+      SAMPLE_HEIGHT: 300
+      SAMPLE_FRAMES: 81
+      PATCH_SIZE: 2
+      PATCH_SIZE_T: 2  # v1.5 diff
+      PATCH_BIAS: False  # v1.5 diff
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 224
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: True 
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+      GRADIENT_CHECKPOINTING: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://ZhipuAI/CogVideoX1.5-5B-I2V@vae/diffusion_pytorch_model.safetensors 
+      SAMPLE_HEIGHT: 768
+      SAMPLE_WIDTH: 1360
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 224
+      CLEAN:
+      USE_GRAD: False
+      T5_DTYPE: bfloat16
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 81
+    IMAGE_SIZE: [768, 1360]
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDataset
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 0
+    NUM_FRAMES: 85
+    FPS: 16
+    HEIGHT: 768
+    WIDTH: 1360
+    PROMPT_PREFIX: 'DISNEY '
+    DATA_TYPE: 'i2v'
+    SAMPLER:
+      NAME: MixtureOfSamplers
+      SUB_SAMPLERS:
+        - NAME: MultiLevelBatchSampler
+          PROB: 1.0
+          FIELDS: [ "video_path", "prompt" ]
+          DELIMITER: '#;#'
+          PATH_PREFIX: cache/datasets/Disney-VideoGeneration-Dataset/
+          INDEX_FILE: cache/datasets/Disney-VideoGeneration-Dataset/index.jsonl
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ "video", "image", "prompt" ]
+        META_KEYS: [ ]
+  #
+#  EVAL_DATA:
+#    NAME: Text2ImageDataset
+#    MODE: eval
+#    PROMPT_FILE:
+#    PROMPT_DATA: [ "A cat running.#;#asset/images/edit_tuner/cat_512.jpg" ]
+#    FIELDS: [ "prompt", "img_path" ]
+#    DELIMITER: '#;#'
+#    PROMPT_PREFIX: ''
+#    PIN_MEMORY: True
+#    BATCH_SIZE: 1
+#    USE_NUM: 8
+#    NUM_WORKERS: 0
+#    IMAGE_SIZE: [768, 1360]
+#    TRANSFORMS:
+#      - NAME: LoadImageFromFileList
+#        FILE_KEYS: [ 'img_path' ]
+#        RGB_ORDER: RGB
+#        BACKEND: pillow
+#      - NAME: FlexibleResize
+#        INTERPOLATION: bilinear
+#        SIZE: [768, 1360]
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'img' ]
+#        BACKEND: pillow
+#      - NAME: FlexibleCenterCrop
+#        SIZE: [768, 1360]
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'img' ]
+#        BACKEND: pillow
+#      - NAME: ImageToTensor
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'img' ]
+#        BACKEND: pillow
+#      - NAME: Normalize
+#        MEAN: [ 0.5,  0.5,  0.5 ]
+#        STD: [ 0.5,  0.5,  0.5 ]
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'image' ]
+#        BACKEND: torchvision
+#      - NAME: Select
+#        KEYS: [ 'image', 'prompt' ]
+#        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+  #
+#  EVAL_HOOKS:
+#    - NAME: ProbeDataHook
+#      PROB_INTERVAL: 100
+#      PRIORITY: 0

scepter/methods/examples/generation/dit_cogvideox1.5_5b_lora.yaml

@@ -0,0 +1,248 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox1.5_5b_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  FPS: 16
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 64
+      LORA_ALPHA: 64
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 0.7
+    INVERT_SCALE_LATENTS: True
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      USE_DYNAMIC_CFG: False
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 1.0  
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: 
+        - ms://ZhipuAI/CogVideoX1.5-5B@transformer/diffusion_pytorch_model-00001-of-00003.safetensors
+        - ms://ZhipuAI/CogVideoX1.5-5B@transformer/diffusion_pytorch_model-00002-of-00003.safetensors
+        - ms://ZhipuAI/CogVideoX1.5-5B@transformer/diffusion_pytorch_model-00003-of-00003.safetensors
+      NUM_ATTENTION_HEADS: 48  
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      NUM_LAYERS: 42  
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 300
+      SAMPLE_HEIGHT: 300
+      SAMPLE_FRAMES: 81
+      PATCH_SIZE: 2
+      PATCH_SIZE_T: 2  # v1.5 diff
+      PATCH_BIAS: False  # v1.5 diff
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 224
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: True 
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+      GRADIENT_CHECKPOINTING: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://ZhipuAI/CogVideoX1.5-5B@vae/diffusion_pytorch_model.safetensors 
+      SAMPLE_HEIGHT: 768
+      SAMPLE_WIDTH: 1360
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 224
+      CLEAN:
+      USE_GRAD: False
+      T5_DTYPE: bfloat16
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 81
+    IMAGE_SIZE: [768, 1360]
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDataset
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 0
+    NUM_FRAMES: 85
+    FPS: 16
+    HEIGHT: 768
+    WIDTH: 1360
+    PROMPT_PREFIX: 'DISNEY '
+    SAMPLER:
+      NAME: MixtureOfSamplers
+      SUB_SAMPLERS:
+        - NAME: MultiLevelBatchSampler
+          PROB: 1.0
+          FIELDS: [ "video_path", "prompt" ]
+          DELIMITER: '#;#'
+          PATH_PREFIX: cache/datasets/Disney-VideoGeneration-Dataset/
+          INDEX_FILE: cache/datasets/Disney-VideoGeneration-Dataset/index.jsonl
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'video', "prompt" ]
+        META_KEYS: [ ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "A girl riding a bike." ]
+    IMAGE_SIZE: [ 768, 1360 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: 'DISNEY ' # ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    USE_NUM: 8
+    NUM_WORKERS: 0
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+  #
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0

scepter/methods/examples/generation/dit_cogvideox_2b_lora.yaml

@@ -0,0 +1,239 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox_2b_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  FPS: 8
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 64
+      LORA_ALPHA: 64
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 1.15258426
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 3.0
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@transformer/diffusion_pytorch_model.safetensors
+      NUM_ATTENTION_HEADS: 30
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      NUM_LAYERS: 30
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 90
+      SAMPLE_HEIGHT: 60
+      SAMPLE_FRAMES: 49
+      PATCH_SIZE: 2
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 226
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: False
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+      GRADIENT_CHECKPOINTING: False
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@vae/diffusion_pytorch_model.safetensors
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: False
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: False
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 226
+      CLEAN:
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 49
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDataset
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    NUM_FRAMES: 49
+    FPS: 8
+    HEIGHT: 480
+    WIDTH: 720
+    PROMPT_PREFIX: 'DISNEY '
+    SAMPLER:
+      NAME: MixtureOfSamplers
+      SUB_SAMPLERS:
+        - NAME: MultiLevelBatchSampler
+          PROB: 1.0
+          FIELDS: [ "video_path", "prompt" ]
+          DELIMITER: '#;#'
+          PATH_PREFIX: cache/datasets/Disney-VideoGeneration-Dataset/
+          INDEX_FILE: cache/datasets/Disney-VideoGeneration-Dataset/index.jsonl
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'video', "prompt" ]
+        META_KEYS: [ ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "A girl riding a bike.", "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical atmosphere of this unique musical performance." ]
+    IMAGE_SIZE: [ 480, 720 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: 'DISNEY '
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    USE_NUM: 8
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+  #
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0

scepter/methods/examples/generation/dit_cogvideox_5b_i2v_lora.yaml

@@ -0,0 +1,270 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox_5b_i2v_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  FPS: 8
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 64
+      LORA_ALPHA: 64
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 0.7  # 5b diff
+    NOISED_IMAGE_DROPOUT: 0.05
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 1.0  # 5b diff
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: # 5b-I2V diff
+        - ms://AI-ModelScope/CogVideoX-5b-I2V@transformer/diffusion_pytorch_model-00001-of-00003.safetensors
+        - ms://AI-ModelScope/CogVideoX-5b-I2V@transformer/diffusion_pytorch_model-00002-of-00003.safetensors
+        - ms://AI-ModelScope/CogVideoX-5b-I2V@transformer/diffusion_pytorch_model-00003-of-00003.safetensors
+      NUM_ATTENTION_HEADS: 48  # 5b diff
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 32  # 5b-I2V diff
+      LATENT_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      NUM_LAYERS: 42  # 5b diff
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 90
+      SAMPLE_HEIGHT: 60
+      SAMPLE_FRAMES: 49
+      PATCH_SIZE: 2
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 226
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: True # 5b diff
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: True  # 5b-I2V diff
+      GRADIENT_CHECKPOINTING: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-5b-I2V@vae/diffusion_pytorch_model.safetensors # 5b diff
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 226
+      CLEAN:
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 49
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDataset
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 0
+    NUM_FRAMES: 49
+    FPS: 8
+    HEIGHT: 480
+    WIDTH: 720
+    PROMPT_PREFIX: 'DISNEY '
+    DATA_TYPE: 'i2v'
+    SAMPLER:
+      NAME: MixtureOfSamplers
+      SUB_SAMPLERS:
+        - NAME: MultiLevelBatchSampler
+          PROB: 1.0
+          FIELDS: [ "video_path", "prompt" ]
+          DELIMITER: '#;#'
+          PATH_PREFIX: cache/datasets/Disney-VideoGeneration-Dataset/
+          INDEX_FILE: cache/datasets/Disney-VideoGeneration-Dataset/index.jsonl
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ "video", "image", "prompt" ]
+        META_KEYS: [ ]
+  #
+#  EVAL_DATA:
+#    NAME: Text2ImageDataset
+#    MODE: eval
+#    PROMPT_FILE:
+#    PROMPT_DATA: [ "A cat running.#;#asset/images/edit_tuner/cat_512.jpg" ]
+#    FIELDS: [ "prompt", "img_path" ]
+#    DELIMITER: '#;#'
+#    PROMPT_PREFIX: ''
+#    PIN_MEMORY: True
+#    BATCH_SIZE: 1
+#    USE_NUM: 8
+#    NUM_WORKERS: 0
+#    IMAGE_SIZE: [ 480, 720 ]
+#    TRANSFORMS:
+#      - NAME: LoadImageFromFileList
+#        FILE_KEYS: [ 'img_path' ]
+#        RGB_ORDER: RGB
+#        BACKEND: pillow
+#      - NAME: FlexibleResize
+#        INTERPOLATION: bilinear
+#        SIZE: [ 480, 720 ]
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'img' ]
+#        BACKEND: pillow
+#      - NAME: FlexibleCenterCrop
+#        SIZE: [ 480, 720 ]
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'img' ]
+#        BACKEND: pillow
+#      - NAME: ImageToTensor
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'img' ]
+#        BACKEND: pillow
+#      - NAME: Normalize
+#        MEAN: [ 0.5,  0.5,  0.5 ]
+#        STD: [ 0.5,  0.5,  0.5 ]
+#        INPUT_KEY: [ 'img' ]
+#        OUTPUT_KEY: [ 'image' ]
+#        BACKEND: torchvision
+#      - NAME: Select
+#        KEYS: [ 'image', 'prompt' ]
+#        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+  #
+#  EVAL_HOOKS:
+#    - NAME: ProbeDataHook
+#      PROB_INTERVAL: 100
+#      PRIORITY: 0

scepter/methods/examples/generation/dit_cogvideox_5b_lora.yaml

@@ -0,0 +1,277 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox_5b_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  FPS: 8
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 64
+      LORA_ALPHA: 64
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 0.7  # 5b diff
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 1.0  # 5b diff
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: # 5b diff
+        - ms://AI-ModelScope/CogVideoX-5b@transformer/diffusion_pytorch_model-00001-of-00002.safetensors
+        - ms://AI-ModelScope/CogVideoX-5b@transformer/diffusion_pytorch_model-00002-of-00002.safetensors
+      NUM_ATTENTION_HEADS: 48  # 5b diff
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      NUM_LAYERS: 42  # 5b diff
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 90
+      SAMPLE_HEIGHT: 60
+      SAMPLE_FRAMES: 49
+      PATCH_SIZE: 2
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 226
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: True # 5b diff
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+      GRADIENT_CHECKPOINTING: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-5b@vae/diffusion_pytorch_model.safetensors # 5b diff
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 226
+      CLEAN:
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 49
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDatasetOTF
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    NUM_FRAMES: 49
+    FPS: 8
+    HEIGHT: 480
+    WIDTH: 720
+    PROMPT_PREFIX: 'DISNEY '
+    DELIMITER: '#;#'
+    FIELDS: [ 'video_path', 'prompt' ]
+    PATH_PREFIX: cache/datasets/Disney-VideoGeneration-Dataset/
+    DATA_FILE: cache/datasets/Disney-VideoGeneration-Dataset/index.jsonl
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'video', 'video_latent', "prompt" ]
+        META_KEYS: [ ]
+    MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-5b@vae/diffusion_pytorch_model.safetensors
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "A girl riding a bike.", "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical atmosphere of this unique musical performance." ]
+    IMAGE_SIZE: [ 480, 720 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: 'DISNEY '
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    USE_NUM: 8
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+  #
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0

scepter/methods/examples/generation/dit_flux1.0_dev_1024_lora.yaml

@@ -0,0 +1,246 @@
+ENV:
+  BACKEND: nccl
+  SEED: 166666
+SOLVER:
+  NAME: LatentDiffusionSolver
+  MAX_STEPS: 100000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_flux_dev_1024_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 16
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.t5_model' ] #
+  SAVE_MODULES: [ 'model'] #
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  FREEZE:
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 4
+      LORA_ALPHA: 4
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "(model.double_blocks.*(.qkv|.proj|.img_mod.lin|.txt_mod.lin))|(model.single_blocks.*(.linear1|.linear2|.modulation.lin))$"
+  ##
+  MODEL:
+    NAME: LatentDiffusionFlux
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: DiffusionFluxRF
+      PREDICTION_TYPE: raw
+      NOISE_SCHEDULER:
+        NAME: FlowMatchSigmaScheduler
+        WEIGHTING_SCHEME: logit_normal
+        SHIFT: 3.0
+        LOGIT_MEAN: 0.0
+        LOGIT_STD: 1.0
+        MODE_SCALE: 1.29
+      SAMPLER_SCHEDULER:
+        NAME: FlowMatchFluxShiftScheduler
+        SHIFT: False
+        SIGMOID_SCALE: 1
+        BASE_SHIFT: 0.5
+        MAX_SHIFT: 1.15
+        #
+    DIFFUSION_MODEL:
+      NAME: Flux
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@flux1-dev.safetensors
+      IN_CHANNELS: 64
+      HIDDEN_SIZE: 3072
+      NUM_HEADS: 24
+      AXES_DIM: [ 16, 56, 56 ]
+      THETA: 10000
+      VEC_IN_DIM: 768
+      GUIDANCE_EMBED: True
+      CONTEXT_IN_DIM: 4096
+      MLP_RATIO: 4.0
+      QKV_BIAS: True
+      DEPTH: 19
+      DEPTH_SINGLE_BLOCKS: 38
+      USE_GRAD_CHECKPOINT: True
+
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLFlux
+      EMBED_DIM: 16
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@ae.safetensors
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 8
+      USE_CONV: False
+      SCALE_FACTOR: 0.3611
+      SHIFT_FACTOR: 0.1159
+      #
+      ENCODER:
+        NAME: Encoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5PlusClipFluxEmbedder
+      T5_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: T5EncoderModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder_2/
+        HF_TOKENIZER_CLS: T5Tokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer_2/
+        MAX_LENGTH: 512
+        OUTPUT_KEY: last_hidden_state
+        D_TYPE: bfloat16
+        BATCH_INFER: False
+        CLEAN: whitespace
+      CLIP_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: CLIPTextModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder/
+        HF_TOKENIZER_CLS: CLIPTokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer/
+        MAX_LENGTH: 77
+        OUTPUT_KEY: pooler_output
+        D_TYPE: bfloat16
+        BATCH_INFER: True
+        CLEAN: whitespace
+        USE_GRAD_CHECKPOINT: True
+  #
+  SAMPLE_ARGS:
+    SAMPLE_STEPS: 50
+    SAMPLER: flow_euler
+    SEED: 2024
+    IMAGE_SIZE: [ 1024, 1024 ]
+    GUIDE_SCALE: 3.5
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 4e-4
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a boy wearing a jacket", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 2
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+#      GRADIENT_CLIP: 1.0
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0

scepter/methods/examples/generation/dit_flux1.0_schnell_1024_lora.yaml

@@ -0,0 +1,267 @@
+ENV:
+  BACKEND: nccl
+  SEED: 166666
+SOLVER:
+  NAME: LatentDiffusionSolver
+  MAX_STEPS: 100000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_flux_schnell_1024_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 16
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.t5_model' ] #
+  SAVE_MODULES: [ 'model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  FREEZE:
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 4
+      LORA_ALPHA: 4
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "(model.double_blocks.*(.qkv|.proj|.img_mod.lin|.txt_mod.lin))|(model.single_blocks.*(.linear1|.linear2|.modulation.lin))$"
+  ##
+  MODEL:
+    NAME: LatentDiffusionFlux
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: DiffusionFluxRF
+      PREDICTION_TYPE: raw
+      NOISE_SCHEDULER:
+        NAME: FlowMatchSigmaScheduler
+        # WEIGHTING_SCHEME DESCRIPTION: The weighting scheme for sampling timesteps, choose from ['sigma_sqrt', 'logit_normal', 'mode', 'cosmap', 'none']. TYPE: str default: 'logit_normal'
+        WEIGHTING_SCHEME: logit_normal
+        SHIFT: 3.0
+        # LOGIT_MEAN DESCRIPTION: The mean of the logit distribution for sampling timesteps. TYPE: float default: 0.0
+        LOGIT_MEAN: 0.0
+        # LOGIT_STD DESCRIPTION: The standard deviation of the logit distribution for sampling timesteps. TYPE: float default: 1.0
+        LOGIT_STD: 1.0
+        # MODE_SCALE DESCRIPTION: The scale factor for the mode of the logit distribution for sampling timesteps. TYPE: float default: 1.29
+        MODE_SCALE: 1.29
+      SAMPLER_SCHEDULER:
+        # NAME DESCRIPTION:  TYPE:  default: 'FlowMatchFluxShiftScheduler'
+        NAME: FlowMatchFluxShiftScheduler
+        # SHIFT DESCRIPTION: Use timestamp shift or not, default is True. TYPE: bool default: True
+        SHIFT: False
+        # SIGMOID_SCALE DESCRIPTION: The scale of sigmoid function for sampling timesteps. TYPE: int default: 1
+        SIGMOID_SCALE: 1
+        # BASE_SHIFT DESCRIPTION: The base shift factor for the timestamp. TYPE: float default: 0.5
+        BASE_SHIFT: 0.5
+        # MAX_SHIFT DESCRIPTION: The max shift factor for the timestamp. TYPE: float default: 1.15
+        MAX_SHIFT: 1.15
+        #
+    DIFFUSION_MODEL:
+      # NAME DESCRIPTION:  TYPE:  default: 'Flux'
+      NAME: Flux
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-schnell@flux1-schnell.safetensors
+      # IN_CHANNELS DESCRIPTION: model's input channels. TYPE: int default: 64
+      IN_CHANNELS: 64
+      # HIDDEN_SIZE DESCRIPTION: model's hidden size. TYPE: int default: 1024
+      HIDDEN_SIZE: 3072
+      # NUM_HEADS DESCRIPTION: number of heads in the transformer. TYPE: int default: 16
+      NUM_HEADS: 24
+      # AXES_DIM DESCRIPTION: dimensions of the axes of the positional encoding. TYPE: list default: [16, 56, 56]
+      AXES_DIM: [ 16, 56, 56 ]
+      # THETA DESCRIPTION: theta for positional encoding. TYPE: int default: 10000
+      THETA: 10000
+      # VEC_IN_DIM DESCRIPTION: dimension of the vector input. TYPE: int default: 768
+      VEC_IN_DIM: 768
+      # GUIDANCE_EMBED DESCRIPTION: whether to use guidance embedding. TYPE: bool default: False
+      GUIDANCE_EMBED: False
+      # CONTEXT_IN_DIM DESCRIPTION: dimension of the context input. TYPE: int default: 4096
+      CONTEXT_IN_DIM: 4096
+      # MLP_RATIO DESCRIPTION: ratio of mlp hidden size to hidden size. TYPE: float default: 4.0
+      MLP_RATIO: 4.0
+      # QKV_BIAS DESCRIPTION: whether to use bias in qkv projection. TYPE: bool default: True
+      QKV_BIAS: True
+      # DEPTH DESCRIPTION: number of transformer blocks. TYPE: int default: 19
+      DEPTH: 19
+      # DEPTH_SINGLE_BLOCKS DESCRIPTION: number of transformer blocks in the single stream block. TYPE: int default: 38
+      DEPTH_SINGLE_BLOCKS: 38
+      USE_GRAD_CHECKPOINT: True
+
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLFlux
+      EMBED_DIM: 16
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-schnell@ae.safetensors
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 8
+      USE_CONV: False
+      SCALE_FACTOR: 0.3611
+      SHIFT_FACTOR: 0.1159
+      #
+      ENCODER:
+        NAME: Encoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5PlusClipFluxEmbedder
+      T5_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: T5EncoderModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-schnell@text_encoder_2/
+        HF_TOKENIZER_CLS: T5Tokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-schnell@tokenizer_2/
+        MAX_LENGTH: 256
+        OUTPUT_KEY: last_hidden_state
+        D_TYPE: bfloat16
+        BATCH_INFER: False
+        CLEAN: whitespace
+      CLIP_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: CLIPTextModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-schnell@text_encoder/
+        HF_TOKENIZER_CLS: CLIPTokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-schnell@tokenizer/
+        MAX_LENGTH: 77
+        OUTPUT_KEY: pooler_output
+        D_TYPE: bfloat16
+        BATCH_INFER: True
+        CLEAN: whitespace
+  #
+  SAMPLE_ARGS:
+    SAMPLE_STEPS: 4
+    SAMPLER: flow_euler
+    SEED: 2024
+    IMAGE_SIZE: [ 1024, 1024 ]
+    GUIDE_SCALE: 3.5
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 4e-4
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a boy wearing a jacket", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 2
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+#      GRADIENT_CLIP: 1.0
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0

scepter/methods/examples/generation/dit_pixart_alpha_1024_lora.yaml

@@ -0,0 +1,223 @@
+ENV:
+  BACKEND: nccl
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 1000
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: 100
+  RESCALE_LR: False
+  #
+  WORK_DIR: ./cache/save_data/dit_pixart_alpha_1024_lora
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  FREEZE:
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 128
+      LORA_ALPHA: 128
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.q|.k|.v|.o|mlp.fc1|mlp.fc2)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionPixart
+    PARAMETERIZATION: eps
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 0.18215
+    SIZE_FACTOR: 8
+    DECODER_BIAS: 0.5
+    DEFAULT_N_PROMPT:
+    SCHEDULE_ARGS:
+      "NAME": "linear"
+      "BETA_MIN": 0.0001
+      "BETA_MAX": 0.02
+    USE_EMA: False
+    LOAD_REFINER: False
+    #
+    DIFFUSION_MODEL:
+      NAME: PixArt
+      PRETRAINED_MODEL: ms://AI-ModelScope/PixArt-alpha@PixArt-XL-2-1024-MS.pth
+      INPUT_SIZE: 128
+      PATCH_SIZE: 2
+      IN_CHANNELS: 4
+      HIDDEN_SIZE: 1152
+      DEPTH: 28
+      NUM_HEADS: 16
+      MLP_RATIO: 4.0
+      CLASS_DROPOUT_PROB: 0.1
+      PRED_SIGMA: True
+      DROP_PATH: 0.0
+      WINDOW_DIZE: 0
+      USE_REL_POS: False
+      CAPTION_CHANNELS: 4096
+      LEWEI_SCALE: 2
+      MODEL_MAX_LENGTH: 120
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-2-base@512-base-ema.safetensors
+      EMBED_DIM: 4
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 1
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/PixArt-alpha@t5-v1_1-xxl/
+      TOKENIZER_PATH: ms://AI-ModelScope/PixArt-alpha@t5-v1_1-xxl/
+      LENGTH: 120
+      CLEAN: heavy
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 20
+    SEED: 2024
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+    DISCRETIZATION: trailing
+    RUN_TRAIN_N: False
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 0.0001
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a boy wearing a jacket", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 0
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 10
+      SHOW_GPU_MEM: True
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/examples/generation/dit_sd3_1024.yaml

@@ -0,0 +1,233 @@
+ENV:
+  BACKEND: nccl
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 500
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: 50
+  RESCALE_LR: False
+  #
+  WORK_DIR: ./cache/save_data/dit_sd3_1024
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    - NAME: "ModelscopeFs"
+      TEMP_DIR: ./cache/cache_data
+  #
+  MODEL:
+    NAME: LatentDiffusionSD3
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 1.5305
+    SHIFT_FACTOR: 0.0609
+    DEFAULT_N_PROMPT:
+    SCHEDULE_ARGS:
+      "NAME": "shifted"
+      "SHIFT": 3
+    USE_EMA: False
+    T_WEIGHT: uniform
+    #
+    DIFFUSION_MODEL:
+      NAME: MMDiT
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+      IGNORE_KEYS: '^first_stage_model.'
+      IN_CHANNELS: 16
+      PATCH_SIZE: 2
+      OUT_CHANNELS: 16
+      DEPTH: 24
+      INPUT_SIZE:
+      ADM_IN_CHANNELS: 2048
+      CONTEXT_EMBEDDER_CONFIG: { 'target': 'torch.nn.Linear', 'params': { 'in_features': 4096, 'out_features': 1536 } }
+      NUM_PATCHES: 36864
+      POS_EMBED_MAX_SIZE: 192
+      POS_EMBED_SCALING_FACTOR:
+      USE_CHECKPOINT: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+      EMBED_DIM: 16
+      IGNORE_KEYS: '^model.diffusion_model.'
+      BATCH_SIZE: 1
+      USE_CONV: False
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: SD3TextEmbedder
+      P_ZERO: 0.0
+      CLIP_L:
+        NAME: FrozenCLIPEmbedder2
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer
+        MAX_LENGTH: 77
+        FREEZE: True
+        LAYER: penultimate
+        RETURN_POOLED: True
+        USE_FINAL_LAYER_NORM: False
+        IS_TRAINABLE: False
+      CLIP_G:
+        NAME: FrozenCLIPEmbedder2
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_2
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_2
+        MAX_LENGTH: 77
+        FREEZE: True
+        LAYER: penultimate
+        RETURN_POOLED: True
+        USE_FINAL_LAYER_NORM: False
+        IS_TRAINABLE: False
+      T5_XXL:
+        NAME: T5EmbedderHF
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_3
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_3
+        LENGTH: 256
+        CLEAN: whitespace
+        USE_GRAD: False
+        T5_DTYPE: float16
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: euler
+    SAMPLE_STEPS: 28
+    SEED: 1749023094
+    GUIDE_SCALE: 5.0
+    GUIDE_RESCALE: 0.0
+    DISCRETIZATION: trailing
+    RUN_TRAIN_N: False
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 1e-5
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a cat holds a blackboard that writes \"hello world\"", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 10000
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 10
+      SHOW_GPU_MEM: True
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 50
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/examples/generation/dit_sd3_1024_lora.yaml

@@ -0,0 +1,241 @@
+ENV:
+  BACKEND: nccl
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 500
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: 50
+  RESCALE_LR: False
+  #
+  WORK_DIR: ./cache/save_data/dit_sd3_1024_lora
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    - NAME: "ModelscopeFs"
+      TEMP_DIR: ./cache/cache_data
+  #
+  TUNER:
+    - NAME: SwiftLoRA
+      R: 128
+      LORA_ALPHA: 128
+      LORA_DROPOUT: 0.0
+      BIAS: "none"
+      TARGET_MODULES: "model.*(.attn.qkv|.attn.proj|mlp.fc1|mlp.fc2)$"
+  #
+  MODEL:
+    NAME: LatentDiffusionSD3
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 1.5305
+    SHIFT_FACTOR: 0.0609
+    DEFAULT_N_PROMPT:
+    SCHEDULE_ARGS:
+      "NAME": "shifted"
+      "SHIFT": 3
+    USE_EMA: False
+    T_WEIGHT: uniform
+    #
+    DIFFUSION_MODEL:
+      NAME: MMDiT
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+      IGNORE_KEYS: '^first_stage_model.'
+      IN_CHANNELS: 16
+      PATCH_SIZE: 2
+      OUT_CHANNELS: 16
+      DEPTH: 24
+      INPUT_SIZE:
+      ADM_IN_CHANNELS: 2048
+      CONTEXT_EMBEDDER_CONFIG: { 'target': 'torch.nn.Linear', 'params': { 'in_features': 4096, 'out_features': 1536 } }
+      NUM_PATCHES: 36864
+      POS_EMBED_MAX_SIZE: 192
+      POS_EMBED_SCALING_FACTOR:
+      USE_CHECKPOINT: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+      EMBED_DIM: 16
+      IGNORE_KEYS: '^model.diffusion_model.'
+      BATCH_SIZE: 1
+      USE_CONV: False
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: SD3TextEmbedder
+      P_ZERO: 0.0
+      CLIP_L:
+        NAME: FrozenCLIPEmbedder2
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer
+        MAX_LENGTH: 77
+        FREEZE: True
+        LAYER: penultimate
+        RETURN_POOLED: True
+        USE_FINAL_LAYER_NORM: False
+        IS_TRAINABLE: False
+      CLIP_G:
+        NAME: FrozenCLIPEmbedder2
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_2
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_2
+        MAX_LENGTH: 77
+        FREEZE: True
+        LAYER: penultimate
+        RETURN_POOLED: True
+        USE_FINAL_LAYER_NORM: False
+        IS_TRAINABLE: False
+      T5_XXL:
+        NAME: T5EmbedderHF
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_3
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_3
+        LENGTH: 256
+        CLEAN: whitespace
+        USE_GRAD: False
+        T5_DTYPE: float16
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: euler
+    SAMPLE_STEPS: 28
+    SEED: 1749023094
+    GUIDE_SCALE: 5.0
+    GUIDE_RESCALE: 0.0
+    DISCRETIZATION: trailing
+    RUN_TRAIN_N: False
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 5e-5
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a cat holds a blackboard that writes \"hello world\"", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 10000
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 10
+      SHOW_GPU_MEM: True
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 50
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/examples/generation/stable_diffusion_1.5_512.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_full
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   MODEL:
     NAME: LatentDiffusion
@@ -124,7 +125,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.0064
+    LEARNING_RATE: 0.00001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -190,7 +191,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_1.5_512_lora.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_lora
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   TUNER:
     -
       NAME: SwiftLoRA
@@ -132,7 +133,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -198,7 +199,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_1.5_512_textlora.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_textlora
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   TUNER:
     -
       NAME: SwiftLoRA
@@ -140,7 +141,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -206,7 +207,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_2.1_512.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_512_full
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   MODEL:
     NAME: LatentDiffusion
@@ -120,7 +121,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.0064
+    LEARNING_RATE: 0.00001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -186,7 +187,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_2.1_512_lora.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_512_lora
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -129,7 +130,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.0064
+    LEARNING_RATE: 0.00001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -195,7 +196,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_2.1_768.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_768_full
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   MODEL:
     NAME: LatentDiffusion
@@ -120,7 +121,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.0064
+    LEARNING_RATE: 0.00001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -186,7 +187,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_2.1_768_lora.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_768_lora
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -129,7 +130,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -195,7 +196,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/examples/generation/stable_diffusion_xl_1024.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_full
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   MODEL:
     NAME: LatentDiffusionXL
@@ -238,7 +239,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.0064
+    LEARNING_RATE: 0.00001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -306,7 +307,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       - NAME: Select

scepter/methods/examples/generation/stable_diffusion_xl_1024_lora.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_lora
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -247,7 +248,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -315,7 +316,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       - NAME: Select

scepter/methods/examples/generation/stable_diffusion_xl_1024_textlora.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_textlora
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -255,7 +256,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -323,7 +324,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       - NAME: Select

scepter/methods/scedit/ctr/sd15_512_sce_ctr_hed.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_sce_ctr_hed
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -141,7 +142,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -167,13 +168,13 @@ SOLVER:
         RGB_ORDER: RGB
         BACKEND: pillow
       - NAME: Resize
-        SIZE: 768
+        SIZE: 512
         INTERPOLATION: bilinear
         INPUT_KEY: [ 'img' ]
         OUTPUT_KEY: [ 'img' ]
         BACKEND: pillow
       - NAME: CenterCrop
-        SIZE: 768
+        SIZE: 512
         INPUT_KEY: [ 'img' ]
         OUTPUT_KEY: [ 'img' ]
         BACKEND: pillow
@@ -215,13 +216,13 @@ SOLVER:
         RGB_ORDER: RGB
         BACKEND: pillow
       - NAME: Resize
-        SIZE: 768
+        SIZE: 512
         INTERPOLATION: bilinear
         INPUT_KEY: [ 'img' ]
         OUTPUT_KEY: [ 'img' ]
         BACKEND: pillow
       - NAME: CenterCrop
-        SIZE: 768
+        SIZE: 512
         INPUT_KEY: [ 'img' ]
         OUTPUT_KEY: [ 'img' ]
         BACKEND: pillow

scepter/methods/scedit/ctr/sd21_768_sce_ctr_canny.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_768_sce_ctr_canny
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -140,7 +141,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/ctr/sd21_768_sce_ctr_pose.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_768_sce_ctr_pose
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -140,7 +141,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/ctr/sdxl_1024_sce_ctr_canny.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_ctr_canny
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -254,7 +255,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/ctr/sdxl_1024_sce_ctr_color.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_ctr_color
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -255,7 +256,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/ctr/sdxl_1024_sce_ctr_color_datatxt.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_ctr_color_datatxt
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -255,7 +256,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/ctr/sdxl_1024_sce_ctr_depth.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_ctr_depth
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -255,7 +256,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/t2i/sd15_512_sce_t2i.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_sce_t2i
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -134,7 +135,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -200,7 +201,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/scedit/t2i/sd15_512_sce_t2i_swift.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_sce_t2i_swift
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -132,7 +133,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -198,7 +199,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/scedit/t2i/sd15_512_textsce_t2i_swift.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd15_512_textsce_t2i_swift
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -140,7 +141,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -206,7 +207,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/scedit/t2i/sd21_768_sce_t2i.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_768_sce_t2i
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -130,7 +131,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -196,7 +197,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/scedit/t2i/sd21_768_sce_t2i_swift.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sd21_768_sce_t2i_swift
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -128,7 +129,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -194,7 +195,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       -

scepter/methods/scedit/t2i/sdxl_1024_sce_t2i.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_t2i
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -247,7 +248,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -315,7 +316,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       - NAME: Select

scepter/methods/scedit/t2i/sdxl_1024_sce_t2i_datatxt.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_t2i_datatxt
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
     FREEZE_PART: [ "first_stage_model", "cond_stage_model", "model" ]
@@ -247,7 +248,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/scedit/t2i/sdxl_1024_sce_t2i_swift.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_sce_t2i_swift
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -245,7 +246,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -313,7 +314,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       - NAME: Select

scepter/methods/scedit/t2i/sdxl_1024_textsce_t2i_swift.yaml

@@ -14,13 +14,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: 100
+  RESCALE_LR: False
   #
   WORK_DIR: ./cache/save_data/sdxl_1024_textsce_t2i_swift
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   TUNER:
     -
@@ -253,7 +254,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -321,7 +322,7 @@ SOLVER:
     NUM_WORKERS: 4
     FILE_SYSTEM:
       NAME: "ModelscopeFs"
-      TEMP_DIR: "./cache/data"
+      TEMP_DIR: "./cache/cache_data"
     #
     TRANSFORMS:
       - NAME: Select

scepter/methods/studio/chatbot/chatbot.yaml

@@ -0,0 +1,25 @@
+WORK_DIR: chatbot
+FILE_SYSTEM:
+  - NAME: LocalFs
+    TEMP_DIR: ./cache/cache_data
+  - NAME: ModelscopeFs
+    TEMP_DIR: ./cache/cache_data
+  - NAME: HuggingfaceFs
+    TEMP_DIR: ./cache/cache_data
+#
+ENABLE_I2V: False
+SKIP_EXAMPLES: True
+#
+MODEL:
+  EDIT_MODEL:
+    MODEL_CFG_DIR: scepter/methods/studio/chatbot/models/
+  I2V:
+    MODEL_NAME: CogVideoX-5b-I2V
+    MODEL_DIR: ms://ZhipuAI/CogVideoX-5b-I2V/
+  CAPTIONER:
+    MODEL_NAME: InternVL2-2B
+    MODEL_DIR: ms://OpenGVLab/InternVL2-2B/
+    PROMPT: '<image>\nThis image is the first frame of a video. Based on this image, please imagine what changes may occur in the next few seconds of the video. Please output brief description, such as "a dog running" or "a person turns to left". No more than 30 words.'
+  ENHANCER:
+    MODEL_NAME: Meta-Llama-3.1-8B-Instruct
+    MODEL_DIR: ms://LLM-Research/Meta-Llama-3.1-8B-Instruct/

scepter/methods/studio/chatbot/models/ace_0.6b_1024.yaml

@@ -0,0 +1,128 @@
+NAME: ACE_0.6B_1024
+IS_DEFAULT: False
+USE_DYNAMIC_MODEL: True
+DEFAULT_PARAS:
+  PARAS:
+  #
+  INPUT:
+    INPUT_IMAGE:
+    INPUT_MASK:
+    TASK:
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    OUTPUT_HEIGHT: 1024
+    OUTPUT_WIDTH: 1024
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+    SEED: -1
+    TAR_INDEX: 0
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        - NAME: encode
+          DTYPE: float16
+          INPUT: ["IMAGE"]
+        - NAME: decode
+          DTYPE: float16
+          INPUT: ["LATENT"]
+    #
+    DIFFUSION_MODEL:
+      FUNCTION:
+        - NAME: forward
+          DTYPE: float16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE"]
+    #
+    COND_STAGE_MODEL:
+      FUNCTION:
+        - NAME: encode_list_of_list
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  NAME: LatentDiffusionACE
+  PRETRAINED_MODEL:
+  IGNORE_KEYS: [ ]
+  SCALE_FACTOR: 0.18215
+  SIZE_FACTOR: 8
+  DECODER_BIAS: 0.5
+  DEFAULT_N_PROMPT: ""
+  TEXT_IDENTIFIER: [ '{image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+  USE_TEXT_POS_EMBEDDINGS: True
+  #
+  DIFFUSION:
+    NAME: BaseDiffusion
+    PREDICTION_TYPE: eps
+    MIN_SNR_GAMMA:
+    NOISE_SCHEDULER:
+      NAME: LinearScheduler
+      NUM_TIMESTEPS: 1000
+      BETA_MIN: 0.0001
+      BETA_MAX: 0.02
+  #
+  DIFFUSION_MODEL:
+    NAME: ACE
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/dit/ace_0.6b_1024px.pth
+    IGNORE_KEYS: [ ]
+    PATCH_SIZE: 2
+    IN_CHANNELS: 4
+    HIDDEN_SIZE: 1152
+    DEPTH: 28
+    NUM_HEADS: 16
+    MLP_RATIO: 4.0
+    PRED_SIGMA: True
+    DROP_PATH: 0.0
+    WINDOW_DIZE: 0
+    Y_CHANNELS: 4096
+    MAX_SEQ_LEN: 4096
+    QK_NORM: True
+    USE_GRAD_CHECKPOINT: True
+    ATTENTION_BACKEND: flash_attn
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKL
+    EMBED_DIM: 4
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/vae/vae.bin
+    IGNORE_KEYS: []
+    #
+    ENCODER:
+      NAME: Encoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    NAME: T5EmbedderHF
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/text_encoder/t5-v1_1-xxl/
+    TOKENIZER_PATH: ms://iic/ACE-0.6B-1024px@models/tokenizer/t5-v1_1-xxl
+    LENGTH: 120
+    T5_DTYPE: bfloat16
+    ADDED_IDENTIFIER: [ '{image}', '{caption}', '{mask}', '{ref_image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+    CLEAN: whitespace
+    USE_GRAD: False

scepter/methods/studio/chatbot/models/ace_0.6b_1024_refiner.yaml

@@ -0,0 +1,284 @@
+NAME: ACE_0.6B_1024_REFINER
+IS_DEFAULT: False
+USE_DYNAMIC_MODEL: True
+DEFAULT_PARAS:
+  PARAS:
+  #
+  INPUT:
+    INPUT_IMAGE:
+    INPUT_MASK:
+    TASK:
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    OUTPUT_HEIGHT: 1024
+    OUTPUT_WIDTH: 1024
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+    SEED: -1
+    TAR_INDEX: 0
+    REFINER_SCALE: 0.2
+    USE_ACE: True
+    #REFINER_PROMPT: "High Resolution, Sharpness, Clarity, Detail Enhancement, Noise Reduction, HD, 4k, Image Restoration, HDR"
+    REFINER_PROMPT: "High Resolution, Sharpness, Clarity, Detail Enhancement, Noise Reduction, HD, 4k, Image Restoration, HDR"
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        - NAME: encode
+          DTYPE: float16
+          INPUT: ["IMAGE"]
+        - NAME: decode
+          DTYPE: float16
+          INPUT: ["LATENT"]
+    #
+    DIFFUSION_MODEL:
+      FUNCTION:
+        - NAME: forward
+          DTYPE: float16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE"]
+    #
+    COND_STAGE_MODEL:
+      FUNCTION:
+        - NAME: encode_list_of_list
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  NAME: LatentDiffusionACE
+  PRETRAINED_MODEL:
+  IGNORE_KEYS: [ ]
+  SCALE_FACTOR: 0.18215
+  SIZE_FACTOR: 8
+  DECODER_BIAS: 0.5
+  DEFAULT_N_PROMPT: ""
+  TEXT_IDENTIFIER: [ '{image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+  USE_TEXT_POS_EMBEDDINGS: True
+  #
+  DIFFUSION:
+    NAME: BaseDiffusion
+    PREDICTION_TYPE: eps
+    MIN_SNR_GAMMA:
+    NOISE_SCHEDULER:
+      NAME: LinearScheduler
+      NUM_TIMESTEPS: 1000
+      BETA_MIN: 0.0001
+      BETA_MAX: 0.02
+  #
+  DIFFUSION_MODEL:
+    NAME: ACE
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/dit/ace_0.6b_1024px.pth
+    IGNORE_KEYS: [ ]
+    PATCH_SIZE: 2
+    IN_CHANNELS: 4
+    HIDDEN_SIZE: 1152
+    DEPTH: 28
+    NUM_HEADS: 16
+    MLP_RATIO: 4.0
+    PRED_SIGMA: True
+    DROP_PATH: 0.0
+    WINDOW_DIZE: 0
+    Y_CHANNELS: 4096
+    MAX_SEQ_LEN: 4096
+    QK_NORM: True
+    USE_GRAD_CHECKPOINT: True
+    ATTENTION_BACKEND: flash_attn
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKL
+    EMBED_DIM: 4
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/vae/vae.bin
+    IGNORE_KEYS: []
+    #
+    ENCODER:
+      NAME: Encoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    NAME: T5EmbedderHF
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-1024px@models/text_encoder/t5-v1_1-xxl/
+    TOKENIZER_PATH: ms://iic/ACE-0.6B-1024px@models/tokenizer/t5-v1_1-xxl
+    LENGTH: 120
+    T5_DTYPE: bfloat16
+    ADDED_IDENTIFIER: [ '{image}', '{caption}', '{mask}', '{ref_image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+    CLEAN: whitespace
+    USE_GRAD: False
+
+ACE_PROMPT: [
+  "A cute cartoon rabbit holding a whiteboard that says 'ACE Refiner', standing in a sunny meadow filled with flowers, with a big smile and bright colors.",
+  "A beautiful young woman with long flowing hair, wearing a summer dress, holding a whiteboard that reads 'ACE Refiner' while sitting on a park bench surrounded by cherry blossoms.",
+  "An adorable cartoon cat wearing oversized glasses, holding a whiteboard that says 'ACE Refiner', perched on a stack of colorful books in a cozy library setting.",
+  "A charming girl with pigtails, wearing a cute school uniform, enthusiastically holding a whiteboard that has 'ACE Refiner' written on it, in a bright and cheerful classroom full of educational posters.",
+  "A friendly cartoon dog with floppy ears, sitting in front of a doghouse, proudly holding a whiteboard that says 'ACE Refiner', with a playful expression and a blue sky in the background.",
+  "A cute anime girl with big expressive eyes, dressed in a colorful outfit, holding a whiteboard that reads 'ACE Refiner' in a fantastical landscape filled with mythical creatures.",
+  "A vibrant cartoon fox holding a whiteboard that says 'ACE Refiner', standing on a rock by a sparkling stream, surrounded by lush greenery and butterflies.",
+  "A stylish young woman in a business outfit, smiling as she holds a whiteboard written with 'ACE Refiner', in a modern office filled with plants and natural light.",
+  "A cute cartoon unicorn holding a sparkling whiteboard that says 'ACE Refiner', frolicking in a magical forest, with rainbows and stars in the background.",
+  "A happy family, consisting of a cute little girl and her playful puppy, holding a whiteboard that says 'ACE Refiner', together in their backyard on a sunny day."
+]
+REFINER_MODEL:
+  NAME: ""
+  IS_DEFAULT: False
+  DEFAULT_PARAS:
+    PARAS:
+      RESOLUTIONS: [ [ 1024, 1024 ] ]
+    INPUT:
+      INPUT_IMAGE:
+      INPUT_MASK:
+      TASK:
+      PROMPT: ""
+      NEGATIVE_PROMPT: ""
+      OUTPUT_HEIGHT: 1024
+      OUTPUT_WIDTH: 1024
+      SAMPLER: flow_euler
+      SAMPLE_STEPS: 30
+      GUIDE_SCALE: 3.5
+      GUIDE_RESCALE:
+    OUTPUT:
+      LATENT:
+      IMAGES:
+      SEED:
+    MODULES_PARAS:
+      FIRST_STAGE_MODEL:
+        FUNCTION:
+          - NAME: encode
+            DTYPE: bfloat16
+            INPUT: [ "IMAGE" ]
+          - NAME: decode
+            DTYPE: bfloat16
+            INPUT: [ "LATENT" ]
+        PARAS:
+          SCALE_FACTOR: 1.5305
+          SHIFT_FACTOR: 0.0609
+          SIZE_FACTOR: 8
+      DIFFUSION_MODEL:
+        FUNCTION:
+          - NAME: forward
+            DTYPE: bfloat16
+            INPUT: [ "SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE" ]
+      COND_STAGE_MODEL:
+        FUNCTION:
+          - NAME: encode
+            DTYPE: bfloat16
+            INPUT: [ "PROMPT" ]
+
+  MODEL:
+    DIFFUSION:
+      NAME: DiffusionFluxRF
+      PREDICTION_TYPE: raw
+      NOISE_SCHEDULER:
+        NAME: FlowMatchSigmaScheduler
+        WEIGHTING_SCHEME: logit_normal
+        SHIFT: 3.0
+        LOGIT_MEAN: 0.0
+        LOGIT_STD: 1.0
+        MODE_SCALE: 1.29
+    DIFFUSION_MODEL:
+      NAME: FluxMR
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@flux1-dev.safetensors
+      IN_CHANNELS: 64
+      OUT_CHANNELS: 64
+      HIDDEN_SIZE: 3072
+      NUM_HEADS: 24
+      AXES_DIM: [ 16, 56, 56 ]
+      THETA: 10000
+      VEC_IN_DIM: 768
+      GUIDANCE_EMBED: True
+      CONTEXT_IN_DIM: 4096
+      MLP_RATIO: 4.0
+      QKV_BIAS: True
+      DEPTH: 19
+      DEPTH_SINGLE_BLOCKS: 38
+      USE_GRAD_CHECKPOINT: True
+      ATTN_BACKEND: flash_attn
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLFlux
+      EMBED_DIM: 16
+      PRETRAINED_MODEL:  ms://AI-ModelScope/FLUX.1-dev@ae.safetensors
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 8
+      USE_CONV: False
+      SCALE_FACTOR: 0.3611
+      SHIFT_FACTOR: 0.1159
+      #
+      ENCODER:
+        NAME: Encoder
+        USE_CHECKPOINT: False
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        USE_CHECKPOINT: False
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5PlusClipFluxEmbedder
+      T5_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: T5EncoderModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder_2/
+        HF_TOKENIZER_CLS: T5Tokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer_2/
+        MAX_LENGTH: 512
+        OUTPUT_KEY: last_hidden_state
+        D_TYPE: bfloat16
+        BATCH_INFER: False
+        CLEAN: whitespace
+      CLIP_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: CLIPTextModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder/
+        HF_TOKENIZER_CLS: CLIPTokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer/
+        MAX_LENGTH: 77
+        OUTPUT_KEY: pooler_output
+        D_TYPE: bfloat16
+        BATCH_INFER: True
+        CLEAN: whitespace

scepter/methods/studio/chatbot/models/ace_0.6b_512.yaml

@@ -0,0 +1,128 @@
+NAME: ACE_0.6B_512
+IS_DEFAULT: True
+USE_DYNAMIC_MODEL: True
+DEFAULT_PARAS:
+  PARAS:
+  #
+  INPUT:
+    INPUT_IMAGE:
+    INPUT_MASK:
+    TASK:
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    OUTPUT_HEIGHT: 512
+    OUTPUT_WIDTH: 512
+    SAMPLER: ddim
+    SAMPLE_STEPS: 20
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+    SEED: -1
+    TAR_INDEX: 0
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        - NAME: encode
+          DTYPE: float16
+          INPUT: ["IMAGE"]
+        - NAME: decode
+          DTYPE: float16
+          INPUT: ["LATENT"]
+    #
+    DIFFUSION_MODEL:
+      FUNCTION:
+        - NAME: forward
+          DTYPE: float16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE"]
+    #
+    COND_STAGE_MODEL:
+      FUNCTION:
+        - NAME: encode_list_of_list
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  NAME: LatentDiffusionACE
+  PRETRAINED_MODEL:
+  IGNORE_KEYS: [ ]
+  SCALE_FACTOR: 0.18215
+  SIZE_FACTOR: 8
+  DECODER_BIAS: 0.5
+  DEFAULT_N_PROMPT: ""
+  TEXT_IDENTIFIER: [ '{image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+  USE_TEXT_POS_EMBEDDINGS: True
+  #
+  DIFFUSION:
+    NAME: BaseDiffusion
+    PREDICTION_TYPE: eps
+    MIN_SNR_GAMMA:
+    NOISE_SCHEDULER:
+      NAME: LinearScheduler
+      NUM_TIMESTEPS: 1000
+      BETA_MIN: 0.0001
+      BETA_MAX: 0.02
+  #
+  DIFFUSION_MODEL:
+    NAME: ACE
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-512px@models/dit/ace_0.6b_512px.pth
+    IGNORE_KEYS: [ ]
+    PATCH_SIZE: 2
+    IN_CHANNELS: 4
+    HIDDEN_SIZE: 1152
+    DEPTH: 28
+    NUM_HEADS: 16
+    MLP_RATIO: 4.0
+    PRED_SIGMA: True
+    DROP_PATH: 0.0
+    WINDOW_DIZE: 0
+    Y_CHANNELS: 4096
+    MAX_SEQ_LEN: 1024
+    QK_NORM: True
+    USE_GRAD_CHECKPOINT: True
+    ATTENTION_BACKEND: flash_attn
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKL
+    EMBED_DIM: 4
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-512px@models/vae/vae.bin
+    IGNORE_KEYS: []
+    #
+    ENCODER:
+      NAME: Encoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    NAME: T5EmbedderHF
+    PRETRAINED_MODEL: ms://iic/ACE-0.6B-512px@models/text_encoder/t5-v1_1-xxl/
+    TOKENIZER_PATH: ms://iic/ACE-0.6B-512px@models/tokenizer/t5-v1_1-xxl
+    LENGTH: 120
+    T5_DTYPE: bfloat16
+    ADDED_IDENTIFIER: [ '{image}', '{caption}', '{mask}', '{ref_image}', '{image1}', '{image2}', '{image3}', '{image4}', '{image5}', '{image6}', '{image7}', '{image8}', '{image9}' ]
+    CLEAN: whitespace
+    USE_GRAD: False

scepter/methods/studio/inference/dit/cogvideox_2b_pro.yaml

@@ -0,0 +1,152 @@
+NAME: COGVIDEOX_2B
+IS_DEFAULT: False
+DEFAULT_PARAS:
+  PARAS:
+    RESOLUTIONS: [[480, 720]]
+  INPUT:
+    IMAGE:
+    ORIGINAL_SIZE_AS_TUPLE: [480, 720]
+    TARGET_SIZE_AS_TUPLE: [480, 720]
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    PROMPT_PREFIX: ""
+    SAMPLE: ddim
+    SAMPLE_STEPS: 50
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    DISCRETIZATION: trailing
+    NUM_FRAMES:
+      DEFAULT: 49
+      VISIBLE: True
+    FPS:
+      DEFAULT: 8
+      VISIBLE: True
+  OUTPUT:
+    VIDEOS:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: decode
+          DTYPE: bfloat16
+          INPUT: ["LATENT"]
+      PARAS:
+          SCALING_FACTOR_IMAGE: 1.15258426
+    DIFFUSION_MODEL:
+      FUNCTION:
+        -
+          NAME: forward
+          DTYPE: bfloat16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE", "GUIDE_RESCALE", "DISCRETIZATION", "NUM_FRAMES", "FPS"]
+      PARAS:
+        USE_ROTARY_POSITIONAL_EMBEDDINGS: False
+        PATCH_SIZE: 2
+        LATENT_CHANNELS: 16
+        SCALE_FACTOR_SPATIAL: 8
+        SCALE_FACTOR_TEMPORAL: 4
+        ATTENTION_HEAD_DIM: 64
+        SAMPLE_HEIGHT: 480
+        SAMPLE_WIDTH: 720
+    COND_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  PRETRAINED_MODEL:
+  DIFFUSION:
+    NAME: BaseDiffusion
+    PREDICTION_TYPE: v
+    NOISE_SCHEDULER:
+      NAME: ScaledLinearScheduler
+      BETA_MIN: 0.00085
+      BETA_MAX: 0.012
+      SNR_SHIFT_SCALE: 3.0
+      RESCALE_BETAS_ZERO_SNR: True
+    DIFFUSION_SAMPLERS:
+      NAME: DDIMSampler
+      DISCRETIZATION_TYPE: trailing
+      ETA: 0.0
+  #
+  DIFFUSION_MODEL:
+    NAME: CogVideoXTransformer3DModel
+    DTYPE: bfloat16
+    PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@transformer/diffusion_pytorch_model.safetensors
+    NUM_ATTENTION_HEADS: 30
+    ATTENTION_HEAD_DIM: 64
+    IN_CHANNELS: 16
+    OUT_CHANNELS: 16
+    FLIP_SIN_TO_COS: True
+    FREQ_SHIFT: 0
+    TIME_EMBED_DIM: 512
+    TEXT_EMBED_DIM: 4096
+    NUM_LAYERS: 30
+    DROPOUT: 0.0
+    ATTENTION_BIAS: True
+    SAMPLE_WIDTH: 90
+    SAMPLE_HEIGHT: 60
+    SAMPLE_FRAMES: 49
+    PATCH_SIZE: 2
+    TEMPORAL_COMPRESSION_RATIO: 4
+    MAX_TEXT_SEQ_LENGTH: 226
+    ACTIVATION_FN: "gelu-approximate"
+    TIMESTEP_ACTIVATION_FN: "silu"
+    NORM_ELEMENTWISE_AFFINE: True
+    NORM_EPS: 1e-5
+    SPATIAL_INTERPOLATION_SCALE: 1.875
+    TEMPORAL_INTERPOLATION_SCALE: 1.0
+    USE_ROTARY_POSITIONAL_EMBEDDINGS: False
+    USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+    GRADIENT_CHECKPOINTING: False
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKLCogVideoX
+    DTYPE: bfloat16
+    PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@vae/diffusion_pytorch_model.safetensors
+    SAMPLE_HEIGHT: 480
+    SAMPLE_WIDTH: 720
+    USE_QUANT_CONV: False
+    USE_POST_QUANT_CONV: False
+    USE_SLICING: True
+    USE_TILING: True
+    GRADIENT_CHECKPOINTING: False
+    ENCODER:
+      NAME: CogVideoXEncoder3D
+      IN_CHANNELS: 3
+      OUT_CHANNELS: 16
+      UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+      BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+      LAYERS_PER_BLOCK: 3
+      ACT_FN: "silu"
+      NORM_EPS: 1e-6
+      NORM_NUM_GROUPS: 32
+      DROPOUT: 0.0
+      PAD_MODE: "first"
+      TEMPORAL_COMPRESSION_RATIO: 4
+      GRADIENT_CHECKPOINTING: False
+    DECODER:
+      NAME: CogVideoXDecoder3D
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 3
+      UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+      BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+      LAYERS_PER_BLOCK: 3
+      ACT_FN: "silu"
+      NORM_EPS: 1e-6
+      NORM_NUM_GROUPS: 32
+      DROPOUT: 0.0
+      PAD_MODE: "first"
+      TEMPORAL_COMPRESSION_RATIO: 4
+      GRADIENT_CHECKPOINTING: False
+  #
+  COND_STAGE_MODEL:
+    NAME: T5EmbedderHF
+    PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+    TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+    LENGTH: 226
+    CLEAN:
+    USE_GRAD: False
+    T5_DTYPE: bfloat16

scepter/methods/studio/inference/dit/cogvideox_5b_pro.yaml

@@ -0,0 +1,154 @@
+NAME: COGVIDEOX_5B
+IS_DEFAULT: False
+DEFAULT_PARAS:
+  PARAS:
+    RESOLUTIONS: [[480, 720]]
+  INPUT:
+    IMAGE:
+    ORIGINAL_SIZE_AS_TUPLE: [480, 720]
+    TARGET_SIZE_AS_TUPLE: [480, 720]
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    PROMPT_PREFIX: ""
+    SAMPLE: ddim
+    SAMPLE_STEPS: 50
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    DISCRETIZATION: trailing
+    NUM_FRAMES:
+      DEFAULT: 49
+      VISIBLE: True
+    FPS:
+      DEFAULT: 8
+      VISIBLE: True
+  OUTPUT:
+    VIDEOS:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: decode
+          DTYPE: bfloat16
+          INPUT: ["LATENT"]
+      PARAS:
+          SCALING_FACTOR_IMAGE: 0.7  # 5b diff
+    DIFFUSION_MODEL:
+      FUNCTION:
+        -
+          NAME: forward
+          DTYPE: bfloat16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE", "GUIDE_RESCALE", "DISCRETIZATION", "NUM_FRAMES", "FPS"]
+      PARAS:
+        USE_ROTARY_POSITIONAL_EMBEDDINGS: True
+        PATCH_SIZE: 2
+        LATENT_CHANNELS: 16
+        SCALE_FACTOR_SPATIAL: 8
+        SCALE_FACTOR_TEMPORAL: 4
+        ATTENTION_HEAD_DIM: 64
+        SAMPLE_HEIGHT: 480
+        SAMPLE_WIDTH: 720
+    COND_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  PRETRAINED_MODEL:
+  DIFFUSION:
+    NAME: BaseDiffusion
+    PREDICTION_TYPE: v
+    NOISE_SCHEDULER:
+      NAME: ScaledLinearScheduler
+      BETA_MIN: 0.00085
+      BETA_MAX: 0.012
+      SNR_SHIFT_SCALE: 1.0  # 5b diff
+      RESCALE_BETAS_ZERO_SNR: True
+    DIFFUSION_SAMPLERS:
+      NAME: DDIMSampler
+      DISCRETIZATION_TYPE: trailing
+      ETA: 0.0
+  #
+  DIFFUSION_MODEL:
+    NAME: CogVideoXTransformer3DModel
+    DTYPE: bfloat16
+    PRETRAINED_MODEL: # 5b diff
+      - ms://AI-ModelScope/CogVideoX-5b@transformer/diffusion_pytorch_model-00001-of-00002.safetensors
+      - ms://AI-ModelScope/CogVideoX-5b@transformer/diffusion_pytorch_model-00002-of-00002.safetensors
+    NUM_ATTENTION_HEADS: 48  # 5b diff
+    ATTENTION_HEAD_DIM: 64
+    IN_CHANNELS: 16
+    OUT_CHANNELS: 16
+    FLIP_SIN_TO_COS: True
+    FREQ_SHIFT: 0
+    TIME_EMBED_DIM: 512
+    TEXT_EMBED_DIM: 4096
+    NUM_LAYERS: 42  # 5b diff
+    DROPOUT: 0.0
+    ATTENTION_BIAS: True
+    SAMPLE_WIDTH: 90
+    SAMPLE_HEIGHT: 60
+    SAMPLE_FRAMES: 49
+    PATCH_SIZE: 2
+    TEMPORAL_COMPRESSION_RATIO: 4
+    MAX_TEXT_SEQ_LENGTH: 226
+    ACTIVATION_FN: "gelu-approximate"
+    TIMESTEP_ACTIVATION_FN: "silu"
+    NORM_ELEMENTWISE_AFFINE: True
+    NORM_EPS: 1e-5
+    SPATIAL_INTERPOLATION_SCALE: 1.875
+    TEMPORAL_INTERPOLATION_SCALE: 1.0
+    USE_ROTARY_POSITIONAL_EMBEDDINGS: True # 5b diff
+    USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+    GRADIENT_CHECKPOINTING: True
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKLCogVideoX
+    DTYPE: bfloat16
+    PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-5b@vae/diffusion_pytorch_model.safetensors # 5b diff
+    SAMPLE_HEIGHT: 480
+    SAMPLE_WIDTH: 720
+    USE_QUANT_CONV: False
+    USE_POST_QUANT_CONV: False
+    USE_SLICING: True
+    USE_TILING: True
+    GRADIENT_CHECKPOINTING: True
+    ENCODER:
+      NAME: CogVideoXEncoder3D
+      IN_CHANNELS: 3
+      OUT_CHANNELS: 16
+      UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+      BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+      LAYERS_PER_BLOCK: 3
+      ACT_FN: "silu"
+      NORM_EPS: 1e-6
+      NORM_NUM_GROUPS: 32
+      DROPOUT: 0.0
+      PAD_MODE: "first"
+      TEMPORAL_COMPRESSION_RATIO: 4
+      GRADIENT_CHECKPOINTING: True
+    DECODER:
+      NAME: CogVideoXDecoder3D
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 3
+      UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+      BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+      LAYERS_PER_BLOCK: 3
+      ACT_FN: "silu"
+      NORM_EPS: 1e-6
+      NORM_NUM_GROUPS: 32
+      DROPOUT: 0.0
+      PAD_MODE: "first"
+      TEMPORAL_COMPRESSION_RATIO: 4
+      GRADIENT_CHECKPOINTING: True
+  #
+  COND_STAGE_MODEL:
+    NAME: T5EmbedderHF
+    PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+    TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+    LENGTH: 226
+    CLEAN:
+    USE_GRAD: False
+    T5_DTYPE: bfloat16

scepter/methods/studio/inference/dit/flux1.0_dev_pro.yaml

@@ -0,0 +1,201 @@
+NAME: FLUX1.0_DEV
+IS_DEFAULT: False
+DEFAULT_PARAS:
+  PARAS:
+    RESOLUTIONS: [[1024, 1024]]
+  INPUT:
+    IMAGE:
+    ORIGINAL_SIZE_AS_TUPLE: [1024, 1024]
+    TARGET_SIZE_AS_TUPLE: [1024, 1024]
+    PROMPT: ""
+    NEGATIVE_PROMPT:
+      DEFAULT: ""
+      VISIBLE: False
+    PROMPT_PREFIX: ""
+    SAMPLE:
+      VALUES: ["flow_euler"]
+      DEFAULT: "flow_euler"
+    SAMPLE_STEPS: 50
+    GUIDE_SCALE: 3.5
+    GUIDE_RESCALE:
+      DEFAULT: 0.0
+      VISIBLE: False
+    DISCRETIZATION:
+      VALUES: []
+      DEFAULT:
+      VISIBLE: False
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: bfloat16
+          INPUT: ["IMAGE"]
+        -
+          NAME: decode
+          DTYPE: bfloat16
+          INPUT: ["LATENT"]
+      PARAS:
+        SCALE_FACTOR: 1.5305
+        SHIFT_FACTOR: 0.0609
+        SIZE_FACTOR: 8
+    DIFFUSION_MODEL:
+      FUNCTION:
+        -
+          NAME: forward
+          DTYPE: bfloat16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE"]
+    COND_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  NAME: LatentDiffusionFlux
+  PARAMETERIZATION: rf
+  TIMESTEPS: 1000
+  MIN_SNR_GAMMA:
+  ZERO_TERMINAL_SNR: False
+  PRETRAINED_MODEL:
+  IGNORE_KEYS: [ ]
+  DEFAULT_N_PROMPT:
+  USE_EMA: False
+  EVAL_EMA: False
+  DIFFUSION:
+    # NAME DESCRIPTION:  TYPE:  default: 'DiffusionFluxRF'
+    NAME: DiffusionFluxRF
+    PREDICTION_TYPE: raw
+    # NOISE_SCHEDULER DESCRIPTION:  TYPE:  default: ''
+    NOISE_SCHEDULER:
+      # NAME DESCRIPTION:  TYPE:  default: 'FlowMatchSigmaScheduler'
+      NAME: FlowMatchSigmaScheduler
+      # WEIGHTING_SCHEME DESCRIPTION: The weighting scheme for sampling timesteps, choose from ['sigma_sqrt', 'logit_normal', 'mode', 'cosmap', 'none']. TYPE: str default: 'logit_normal'
+      WEIGHTING_SCHEME: logit_normal
+      SHIFT: 3.0
+      # LOGIT_MEAN DESCRIPTION: The mean of the logit distribution for sampling timesteps. TYPE: float default: 0.0
+      LOGIT_MEAN: 0.0
+      # LOGIT_STD DESCRIPTION: The standard deviation of the logit distribution for sampling timesteps. TYPE: float default: 1.0
+      LOGIT_STD: 1.0
+      # MODE_SCALE DESCRIPTION: The scale factor for the mode of the logit distribution for sampling timesteps. TYPE: float default: 1.29
+      MODE_SCALE: 1.29
+      #
+  DIFFUSION_MODEL:
+    # NAME DESCRIPTION:  TYPE:  default: 'Flux'
+    NAME: Flux
+    PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@flux1-dev.safetensors
+    # IN_CHANNELS DESCRIPTION: model's input channels. TYPE: int default: 64
+    IN_CHANNELS: 64
+    # HIDDEN_SIZE DESCRIPTION: model's hidden size. TYPE: int default: 1024
+    HIDDEN_SIZE: 3072
+    # NUM_HEADS DESCRIPTION: number of heads in the transformer. TYPE: int default: 16
+    NUM_HEADS: 24
+    # AXES_DIM DESCRIPTION: dimensions of the axes of the positional encoding. TYPE: list default: [16, 56, 56]
+    AXES_DIM: [ 16, 56, 56 ]
+    # THETA DESCRIPTION: theta for positional encoding. TYPE: int default: 10000
+    THETA: 10000
+    # VEC_IN_DIM DESCRIPTION: dimension of the vector input. TYPE: int default: 768
+    VEC_IN_DIM: 768
+    # GUIDANCE_EMBED DESCRIPTION: whether to use guidance embedding. TYPE: bool default: False
+    GUIDANCE_EMBED: True
+    # CONTEXT_IN_DIM DESCRIPTION: dimension of the context input. TYPE: int default: 4096
+    CONTEXT_IN_DIM: 4096
+    # MLP_RATIO DESCRIPTION: ratio of mlp hidden size to hidden size. TYPE: float default: 4.0
+    MLP_RATIO: 4.0
+    # QKV_BIAS DESCRIPTION: whether to use bias in qkv projection. TYPE: bool default: True
+    QKV_BIAS: True
+    # DEPTH DESCRIPTION: number of transformer blocks. TYPE: int default: 19
+    DEPTH: 19
+    # DEPTH_SINGLE_BLOCKS DESCRIPTION: number of transformer blocks in the single stream block. TYPE: int default: 38
+    DEPTH_SINGLE_BLOCKS: 38
+
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKLFlux
+    EMBED_DIM: 16
+    PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@ae.safetensors
+    IGNORE_KEYS: [ ]
+    BATCH_SIZE: 8
+    USE_CONV: False
+    SCALE_FACTOR: 0.3611
+    SHIFT_FACTOR: 0.1159
+    #
+    ENCODER:
+      NAME: Encoder
+      USE_CHECKPOINT: True
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 16
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      USE_CHECKPOINT: True
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 16
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    # NAME DESCRIPTION:  TYPE:  default: 'T5PlusClipFluxEmbedder'
+    NAME: T5PlusClipFluxEmbedder
+    # T5_MODEL DESCRIPTION:  TYPE:  default: ''
+    T5_MODEL:
+      # NAME DESCRIPTION:  TYPE:  default: 'HFEmbedder'
+      NAME: HFEmbedder
+      # HF_MODEL_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_MODEL_CLS: T5EncoderModel
+      # MODEL_PATH DESCRIPTION: model folder path TYPE: NoneType default: None
+      MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder_2/
+      # HF_TOKENIZER_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_TOKENIZER_CLS: T5Tokenizer
+      # TOKENIZER_PATH DESCRIPTION: tokenizer folder path TYPE: NoneType default: None
+      TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer_2/
+      # MAX_LENGTH DESCRIPTION: max length of input TYPE: int default: 77
+      MAX_LENGTH: 512
+      # OUTPUT_KEY DESCRIPTION: output key TYPE: str default: 'last_hidden_state'
+      OUTPUT_KEY: last_hidden_state
+      # D_TYPE DESCRIPTION: dtype TYPE: str default: 'bfloat16'
+      D_TYPE: bfloat16
+      # BATCH_INFER DESCRIPTION: batch infer TYPE: bool default: False
+      BATCH_INFER: False
+      CLEAN: whitespace
+    # CLIP_MODEL DESCRIPTION:  TYPE:  default: ''
+    CLIP_MODEL:
+      # NAME DESCRIPTION:  TYPE:  default: 'HFEmbedder'
+      NAME: HFEmbedder
+      # HF_MODEL_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_MODEL_CLS: CLIPTextModel
+      # MODEL_PATH DESCRIPTION: model folder path TYPE: NoneType default: None
+      MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder/
+      # HF_TOKENIZER_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_TOKENIZER_CLS: CLIPTokenizer
+      # TOKENIZER_PATH DESCRIPTION: tokenizer folder path TYPE: NoneType default: None
+      TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer/
+      # MAX_LENGTH DESCRIPTION: max length of input TYPE: int default: 77
+      MAX_LENGTH: 77
+      # OUTPUT_KEY DESCRIPTION: output key TYPE: str default: 'last_hidden_state'
+      OUTPUT_KEY: pooler_output
+      # D_TYPE DESCRIPTION: dtype TYPE: str default: 'bfloat16'
+      D_TYPE: bfloat16
+      # BATCH_INFER DESCRIPTION: batch infer TYPE: bool default: False
+      BATCH_INFER: True
+      CLEAN: whitespace

scepter/methods/studio/inference/dit/flux1.0_schnell_pro.yaml

@@ -0,0 +1,212 @@
+NAME: FLUX1.0_SCHNELL
+IS_DEFAULT: False
+DEFAULT_PARAS:
+  PARAS:
+    RESOLUTIONS: [[1024, 1024]]
+  INPUT:
+    IMAGE:
+    ORIGINAL_SIZE_AS_TUPLE: [1024, 1024]
+    TARGET_SIZE_AS_TUPLE: [1024, 1024]
+    PROMPT: ""
+    NEGATIVE_PROMPT:
+      DEFAULT: ""
+      VISIBLE: False
+    PROMPT_PREFIX: ""
+    SAMPLE:
+      VALUES: ["flow_euler"]
+      DEFAULT: "flow_euler"
+    SAMPLE_STEPS: 4
+    GUIDE_SCALE: 3.5
+    GUIDE_RESCALE:
+      DEFAULT: 0.0
+      VISIBLE: False
+    DISCRETIZATION:
+      VALUES: []
+      DEFAULT:
+      VISIBLE: False
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: bfloat16
+          INPUT: ["IMAGE"]
+        -
+          NAME: decode
+          DTYPE: bfloat16
+          INPUT: ["LATENT"]
+      PARAS:
+        SCALE_FACTOR: 1.5305
+        SHIFT_FACTOR: 0.0609
+        SIZE_FACTOR: 8
+    DIFFUSION_MODEL:
+      FUNCTION:
+        -
+          NAME: forward
+          DTYPE: bfloat16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE"]
+    COND_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: bfloat16
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  NAME: LatentDiffusionFlux
+  PARAMETERIZATION: rf
+  TIMESTEPS: 1000
+  MIN_SNR_GAMMA:
+  ZERO_TERMINAL_SNR: False
+  PRETRAINED_MODEL:
+  IGNORE_KEYS: [ ]
+  DEFAULT_N_PROMPT:
+  USE_EMA: False
+  EVAL_EMA: False
+  DIFFUSION:
+    # NAME DESCRIPTION:  TYPE:  default: 'DiffusionFluxRF'
+    NAME: DiffusionFluxRF
+    PREDICTION_TYPE: raw
+    # NOISE_SCHEDULER DESCRIPTION:  TYPE:  default: ''
+    NOISE_SCHEDULER:
+      # NAME DESCRIPTION:  TYPE:  default: 'FlowMatchSigmaScheduler'
+      NAME: FlowMatchSigmaScheduler
+      # WEIGHTING_SCHEME DESCRIPTION: The weighting scheme for sampling timesteps, choose from ['sigma_sqrt', 'logit_normal', 'mode', 'cosmap', 'none']. TYPE: str default: 'logit_normal'
+      WEIGHTING_SCHEME: logit_normal
+      SHIFT: 3.0
+      # LOGIT_MEAN DESCRIPTION: The mean of the logit distribution for sampling timesteps. TYPE: float default: 0.0
+      LOGIT_MEAN: 0.0
+      # LOGIT_STD DESCRIPTION: The standard deviation of the logit distribution for sampling timesteps. TYPE: float default: 1.0
+      LOGIT_STD: 1.0
+      # MODE_SCALE DESCRIPTION: The scale factor for the mode of the logit distribution for sampling timesteps. TYPE: float default: 1.29
+      MODE_SCALE: 1.29
+    SAMPLER_SCHEDULER:
+      # NAME DESCRIPTION:  TYPE:  default: 'FlowMatchFluxShiftScheduler'
+      NAME: FlowMatchFluxShiftScheduler
+      # SHIFT DESCRIPTION: Use timestamp shift or not, default is True. TYPE: bool default: True
+      SHIFT: False
+      # SIGMOID_SCALE DESCRIPTION: The scale of sigmoid function for sampling timesteps. TYPE: int default: 1
+      SIGMOID_SCALE: 1
+      # BASE_SHIFT DESCRIPTION: The base shift factor for the timestamp. TYPE: float default: 0.5
+      BASE_SHIFT: 0.5
+      # MAX_SHIFT DESCRIPTION: The max shift factor for the timestamp. TYPE: float default: 1.15
+      MAX_SHIFT: 1.15
+      #
+  DIFFUSION_MODEL:
+    # NAME DESCRIPTION:  TYPE:  default: 'Flux'
+    NAME: Flux
+    PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-schnell@flux1-schnell.safetensors
+    # IN_CHANNELS DESCRIPTION: model's input channels. TYPE: int default: 64
+    IN_CHANNELS: 64
+    # HIDDEN_SIZE DESCRIPTION: model's hidden size. TYPE: int default: 1024
+    HIDDEN_SIZE: 3072
+    # NUM_HEADS DESCRIPTION: number of heads in the transformer. TYPE: int default: 16
+    NUM_HEADS: 24
+    # AXES_DIM DESCRIPTION: dimensions of the axes of the positional encoding. TYPE: list default: [16, 56, 56]
+    AXES_DIM: [ 16, 56, 56 ]
+    # THETA DESCRIPTION: theta for positional encoding. TYPE: int default: 10000
+    THETA: 10000
+    # VEC_IN_DIM DESCRIPTION: dimension of the vector input. TYPE: int default: 768
+    VEC_IN_DIM: 768
+    # GUIDANCE_EMBED DESCRIPTION: whether to use guidance embedding. TYPE: bool default: False
+    GUIDANCE_EMBED: False
+    # CONTEXT_IN_DIM DESCRIPTION: dimension of the context input. TYPE: int default: 4096
+    CONTEXT_IN_DIM: 4096
+    # MLP_RATIO DESCRIPTION: ratio of mlp hidden size to hidden size. TYPE: float default: 4.0
+    MLP_RATIO: 4.0
+    # QKV_BIAS DESCRIPTION: whether to use bias in qkv projection. TYPE: bool default: True
+    QKV_BIAS: True
+    # DEPTH DESCRIPTION: number of transformer blocks. TYPE: int default: 19
+    DEPTH: 19
+    # DEPTH_SINGLE_BLOCKS DESCRIPTION: number of transformer blocks in the single stream block. TYPE: int default: 38
+    DEPTH_SINGLE_BLOCKS: 38
+
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKLFlux
+    EMBED_DIM: 16
+    PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-schnell@ae.safetensors
+    IGNORE_KEYS: [ ]
+    BATCH_SIZE: 8
+    USE_CONV: False
+    SCALE_FACTOR: 0.3611
+    SHIFT_FACTOR: 0.1159
+    #
+    ENCODER:
+      NAME: Encoder
+      USE_CHECKPOINT: True
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 16
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      USE_CHECKPOINT: True
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 16
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    # NAME DESCRIPTION:  TYPE:  default: 'T5PlusClipFluxEmbedder'
+    NAME: T5PlusClipFluxEmbedder
+    # T5_MODEL DESCRIPTION:  TYPE:  default: ''
+    T5_MODEL:
+      # NAME DESCRIPTION:  TYPE:  default: 'HFEmbedder'
+      NAME: HFEmbedder
+      # HF_MODEL_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_MODEL_CLS: T5EncoderModel
+      # MODEL_PATH DESCRIPTION: model folder path TYPE: NoneType default: None
+      MODEL_PATH: ms://AI-ModelScope/FLUX.1-schnell@text_encoder_2/
+      # HF_TOKENIZER_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_TOKENIZER_CLS: T5Tokenizer
+      # TOKENIZER_PATH DESCRIPTION: tokenizer folder path TYPE: NoneType default: None
+      TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-schnell@tokenizer_2/
+      # MAX_LENGTH DESCRIPTION: max length of input TYPE: int default: 77
+      MAX_LENGTH: 256
+      # OUTPUT_KEY DESCRIPTION: output key TYPE: str default: 'last_hidden_state'
+      OUTPUT_KEY: last_hidden_state
+      # D_TYPE DESCRIPTION: dtype TYPE: str default: 'bfloat16'
+      D_TYPE: bfloat16
+      # BATCH_INFER DESCRIPTION: batch infer TYPE: bool default: False
+      BATCH_INFER: False
+      CLEAN: whitespace
+    # CLIP_MODEL DESCRIPTION:  TYPE:  default: ''
+    CLIP_MODEL:
+      # NAME DESCRIPTION:  TYPE:  default: 'HFEmbedder'
+      NAME: HFEmbedder
+      # HF_MODEL_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_MODEL_CLS: CLIPTextModel
+      # MODEL_PATH DESCRIPTION: model folder path TYPE: NoneType default: None
+      MODEL_PATH: ms://AI-ModelScope/FLUX.1-schnell@text_encoder/
+      # HF_TOKENIZER_CLS DESCRIPTION: huggingface cls in transfomer TYPE: NoneType default: None
+      HF_TOKENIZER_CLS: CLIPTokenizer
+      # TOKENIZER_PATH DESCRIPTION: tokenizer folder path TYPE: NoneType default: None
+      TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-schnell@tokenizer/
+      # MAX_LENGTH DESCRIPTION: max length of input TYPE: int default: 77
+      MAX_LENGTH: 77
+      # OUTPUT_KEY DESCRIPTION: output key TYPE: str default: 'last_hidden_state'
+      OUTPUT_KEY: pooler_output
+      # D_TYPE DESCRIPTION: dtype TYPE: str default: 'bfloat16'
+      D_TYPE: bfloat16
+      # BATCH_INFER DESCRIPTION: batch infer TYPE: bool default: False
+      BATCH_INFER: True
+      CLEAN: whitespace

scepter/methods/studio/inference/dit/pixart_alpha_pro.yaml

@@ -0,0 +1,120 @@
+NAME: PIXART_ALPHA
+IS_DEFAULT: False
+DEFAULT_PARAS:
+  PARAS:
+    RESOLUTIONS: [[1024, 1024]]
+  INPUT:
+    IMAGE:
+    ORIGINAL_SIZE_AS_TUPLE: [1024, 1024]
+    TARGET_SIZE_AS_TUPLE: [1024, 1024]
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    PROMPT_PREFIX: ""
+    SAMPLE: ddim
+    SAMPLE_STEPS: 20
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+    DISCRETIZATION: trailing
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: float32
+          INPUT: ["IMAGE"]
+        -
+          NAME: decode
+          DTYPE: float32
+          INPUT: ["LATENT"]
+      PARAS:
+        SCALE_FACTOR: 0.18215
+        SIZE_FACTOR: 8
+    DIFFUSION_MODEL:
+      FUNCTION:
+        -
+          NAME: forward
+          DTYPE: float16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE", "GUIDE_RESCALE", "DISCRETIZATION"]
+    COND_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: float32
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  PRETRAINED_MODEL:
+  DECODER_BIAS: 0.5
+  SCHEDULE:
+    PARAMETERIZATION: "eps"
+    TIMESTEPS: 1000
+    ZERO_TERMINAL_SNR: False
+    SCHEDULE_ARGS:
+      "NAME": "linear"
+      "BETA_MIN": 0.0001
+      "BETA_MAX": 0.02
+  #
+  DIFFUSION_MODEL:
+    NAME: PixArt
+    PRETRAINED_MODEL: ms://AI-ModelScope/PixArt-alpha@PixArt-XL-2-1024-MS.pth
+    INPUT_SIZE: 128
+    PATCH_SIZE: 2
+    IN_CHANNELS: 4
+    HIDDEN_SIZE: 1152
+    DEPTH: 28
+    NUM_HEADS: 16
+    MLP_RATIO: 4.0
+    CLASS_DROPOUT_PROB: 0.1
+    PRED_SIGMA: True
+    DROP_PATH: 0.0
+    WINDOW_DIZE: 0
+    USE_REL_POS: False
+    CAPTION_CHANNELS: 4096
+    LEWEI_SCALE: 2
+    MODEL_MAX_LENGTH: 120
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKL
+    PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-2-base@512-base-ema.safetensors
+    EMBED_DIM: 4
+    IGNORE_KEYS: [ ]
+    BATCH_SIZE: 1
+    #
+    ENCODER:
+      NAME: Encoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 4
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    NAME: T5EmbedderHF
+    PRETRAINED_MODEL: ms://AI-ModelScope/PixArt-alpha@t5-v1_1-xxl/
+    TOKENIZER_PATH: ms://AI-ModelScope/PixArt-alpha@t5-v1_1-xxl/
+    LENGTH: 120
+    CLEAN: heavy
+    USE_GRAD: False

scepter/methods/studio/inference/dit/sd3_pro.yaml

@@ -0,0 +1,148 @@
+NAME: SD3
+IS_DEFAULT: False
+DEFAULT_PARAS:
+  PARAS:
+    RESOLUTIONS: [[1024, 1024]]
+  INPUT:
+    IMAGE:
+    ORIGINAL_SIZE_AS_TUPLE: [1024, 1024]
+    TARGET_SIZE_AS_TUPLE: [1024, 1024]
+    PROMPT: ""
+    NEGATIVE_PROMPT: ""
+    PROMPT_PREFIX: ""
+    SAMPLE: euler
+    SAMPLE_STEPS: 28
+    GUIDE_SCALE: 5.0
+    GUIDE_RESCALE: 0.0
+    DISCRETIZATION: trailing
+  OUTPUT:
+    LATENT:
+    IMAGES:
+    SEED:
+  MODULES_PARAS:
+    FIRST_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: float32
+          INPUT: ["IMAGE"]
+        -
+          NAME: decode
+          DTYPE: float32
+          INPUT: ["LATENT"]
+      PARAS:
+        SCALE_FACTOR: 1.5305
+        SHIFT_FACTOR: 0.0609
+        SIZE_FACTOR: 8
+    DIFFUSION_MODEL:
+      FUNCTION:
+        -
+          NAME: forward
+          DTYPE: float16
+          INPUT: ["SAMPLE_STEPS", "SAMPLE", "GUIDE_SCALE", "GUIDE_RESCALE", "DISCRETIZATION"]
+    COND_STAGE_MODEL:
+      FUNCTION:
+        -
+          NAME: encode
+          DTYPE: float32
+          INPUT: ["PROMPT"]
+#
+MODEL:
+  PRETRAINED_MODEL:
+  SCHEDULE:
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 1.5305
+    SHIFT_FACTOR: 0.0609
+    DEFAULT_N_PROMPT:
+    SCHEDULE_ARGS:
+      "NAME": "shifted"
+      "SHIFT": 3
+    T_WEIGHT: uniform
+  #
+  DIFFUSION_MODEL:
+    NAME: MMDiT
+    PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+    IGNORE_KEYS: '^first_stage_model.'
+    IN_CHANNELS: 16
+    PATCH_SIZE: 2
+    OUT_CHANNELS: 16
+    DEPTH: 24
+    INPUT_SIZE:
+    ADM_IN_CHANNELS: 2048
+    CONTEXT_EMBEDDER_CONFIG: { 'target': 'torch.nn.Linear', 'params': { 'in_features': 4096, 'out_features': 1536 } }
+    NUM_PATCHES: 36864
+    POS_EMBED_MAX_SIZE: 192
+    POS_EMBED_SCALING_FACTOR:
+    USE_CHECKPOINT: True
+  #
+  FIRST_STAGE_MODEL:
+    NAME: AutoencoderKL
+    PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+    EMBED_DIM: 16
+    IGNORE_KEYS: '^model.diffusion_model.'
+    BATCH_SIZE: 1
+    USE_CONV: False
+    #
+    ENCODER:
+      NAME: Encoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 16
+      DOUBLE_Z: True
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+    #
+    DECODER:
+      NAME: Decoder
+      CH: 128
+      OUT_CH: 3
+      NUM_RES_BLOCKS: 2
+      IN_CHANNELS: 3
+      ATTN_RESOLUTIONS: [ ]
+      CH_MULT: [ 1, 2, 4, 4 ]
+      Z_CHANNELS: 16
+      DROPOUT: 0.0
+      RESAMP_WITH_CONV: True
+      GIVE_PRE_END: False
+      TANH_OUT: False
+  #
+  COND_STAGE_MODEL:
+    NAME: SD3TextEmbedder
+    P_ZERO: 0.0
+    CLIP_L:
+      NAME: FrozenCLIPEmbedder2
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder
+      TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer
+      MAX_LENGTH: 77
+      FREEZE: True
+      LAYER: penultimate
+      RETURN_POOLED: True
+      USE_FINAL_LAYER_NORM: False
+      IS_TRAINABLE: False
+    CLIP_G:
+      NAME: FrozenCLIPEmbedder2
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_2
+      TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_2
+      MAX_LENGTH: 77
+      FREEZE: True
+      LAYER: penultimate
+      RETURN_POOLED: True
+      USE_FINAL_LAYER_NORM: False
+      IS_TRAINABLE: False
+    T5_XXL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_3
+      TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_3
+      LENGTH: 256
+      CLEAN: whitespace
+      USE_GRAD: False
+      T5_DTYPE: float16

scepter/methods/studio/inference/inference.yaml

@@ -1,4 +1,5 @@
 WORK_DIR: "inference"
+SKIP_EXAMPLES: True
 DIFFUSION_PARAS:
   SAMPLE:
     VALUES: ['ddim', 'euler', 'euler_ancestral', 'heun', 'dpm2',
@@ -6,65 +7,91 @@ DIFFUSION_PARAS:
              'dpm2_karras', 'dpm2_ancestral_karras', 'dpmpp_2s_ancestral_karras', 'dpmpp_2m_karras',
              'dpmpp_sde_karras', 'dpmpp_2m_sde_karras']
     DEFAULT: 'dpmpp_2s_ancestral'
+    VISIBLE: True
   NEGATIVE_PROMPT:
     DEFAULT:
+    VISIBLE: True
   PROMPT_PREFIX:
     DEFAULT:
+    VISIBLE: True
   SAMPLES:
     MIN: 1
     MAX: 4
     DEFAULT: 1
+    VISIBLE: True
+  NUM_FRAMES:
+    MIN: 1
+    MAX: 100
+    DEFAULT: 49
+    VISIBLE: False
+  FPS:
+    MIN: 1
+    MAX: 50
+    DEFAULT: 8
+    VISIBLE: False
   SAMPLE_STEPS:
     MIN: 1
     MAX: 100
     DEFAULT: 30
+    VISIBLE: True
   GUIDE_SCALE:
     MIN: 0
     MAX: 10
     DEFAULT: 5.0
+    VISIBLE: True
   GUIDE_RESCALE:
     MIN: 0
     MAX: 1.0
     DEFAULT: 0.5
+    VISIBLE: True
   DISCRETIZATION:
     VALUES: ["trailing", "leading", "linspace"]
     DEFAULT: "linspace"
+    VISIBLE: True
   REFINE_SAMPLERS:
     VALUES: [ 'ddim', 'euler', 'euler_ancestral', 'heun', 'dpm2',
               'dpm2_ancestral', 'dpmpp_2m', 'dpmpp_sde', 'dpmpp_2m_sde', 'dpmpp_2s_ancestral',
               'dpm2_karras', 'dpm2_ancestral_karras', 'dpmpp_2s_ancestral_karras', 'dpmpp_2m_karras',
               'dpmpp_sde_karras', 'dpmpp_2m_sde_karras' ]
     DEFAULT: 'dpmpp_2s_ancestral'
+    VISIBLE: True
   REFINE_SAMPLE_STEPS:
     MIN: 0
     MAX: 100
     DEFAULT: 30
+    VISIBLE: True
   REFINE_GUIDE_SCALE:
     MIN: 0
     MAX: 10
     DEFAULT: 5.0
+    VISIBLE: True
   REFINE_GUIDE_RESCALE:
     MIN: 0
     MAX: 1.0
     DEFAULT: 0.5
+    VISIBLE: True
   REFINE_DISCRETIZATION:
     VALUES: [ "trailing", "leading", "linspace" ]
     DEFAULT: "linspace"
+    VISIBLE: True
   AESTHETIC_SCORE:
     MIN: 0.0
     MAX: 10.0
     DEFAULT: 6.0
+    VISIBLE: True
   NEGATIVE_AESTHETIC_SCORE:
     MIN: 0.0
     MAX: 10.0
     DEFAULT: 2.5
+    VISIBLE: True
   REFINE_STRENGTH:
     MIN: 0
     MAX: 1.0
     DEFAULT: 0.15
   RESOLUTIONS:
-    VALUES: [ [512, 512], [768, 768],
-              [704, 1408], [704, 1344], [768, 1344],
+    VALUES: [
+             [512, 512], [768, 768],
+             [704, 1408], [704, 1344], [768, 1344],
              [720, 1280],
              [768, 1280], [832, 1216], [832, 1152],
              [896, 1152], [896, 1088], [960, 1088],
@@ -74,8 +101,14 @@ DIFFUSION_PARAS:
              [1280, 768],
              [1344, 768], [1344, 704], [1408, 704],
              [1472, 704], [1536, 640], [1600, 640],
-             [1664, 576], [1728, 576]]
+             [1664, 576], [1728, 576],
+             [2048, 2048], [2048, 1920], [1920, 2048],
+             [1536, 2560], [2560, 1536], [2560, 1440],
+             [2560, 1440],
+             [480, 720], [720, 480]
+            ]
     DEFAULT: [1024, 1024]
+    VISIBLE: True
 EXTENSION_PARAS:
   MANTRA_BOOK: scepter/methods/studio/extensions/mantra_book/mantra_book.yaml
   OFFICIAL_TUNERS: scepter/methods/studio/extensions/tuners/official_tuners.yaml

scepter/methods/studio/preprocess/preprocess.yaml

@@ -2,11 +2,10 @@ WORK_DIR: datasets
 EXPORT_DIR: export_datasets
 FILE_SYSTEM:
   -
-    # NAME DESCRIPTION:  TYPE:  default: ''
     NAME: LocalFs
     AUTO_CLEAN: False
-
 PROCESSORS:
+  # Caption processor
   - NAME: BlipImageBase
     TYPE: caption
     MODEL_PATH: ms://cubeai/blip-image-captioning-base
@@ -15,6 +14,81 @@ PROCESSORS:
     PARAS:
       - LANGUAGE_NAME: English
         LANGUAGE_ZH_NAME: 英语
+  - NAME: InternVL15
+    TYPE: caption
+    MODEL_PATH: ms://AI-ModelScope/InternVL-Chat-V1-5
+    DEVICE: "gpu"
+    MEMORY: 49968
+    PARAS:
+      - PROMPT: 用中文描述这张图片
+        LANGUAGE_NAME: Chinese
+        LANGUAGE_ZH_NAME: 中文
+      - PROMPT: Generate the caption in English
+        LANGUAGE_NAME: English
+        LANGUAGE_ZH_NAME: 英语
+  - NAME: QWVLQuantize
+    TYPE: caption
+    DEVICE: "gpu"
+    MEMORY: 7885
+    MODEL_PATH: ms://qwen/Qwen-VL:v1.0.3
+    PARAS:
+      - PROMPT: 用中文描述这张图片
+        LANGUAGE_NAME: Chinese
+        LANGUAGE_ZH_NAME: 中文
+        MAX_NEW_TOKENS:
+          VALUE: 1024
+          MAX: 2048
+          STEP: 128
+          MIN: 256
+        MIN_NEW_TOKENS:
+          VALUE: 16
+          MAX: 1024
+          STEP: 16
+          MIN: 0
+        NUM_BEAMS:
+          VALUE: 1
+          MAX: 12
+          STEP: 1
+          MIN: 1
+        REPETITION_PENALTY:
+          VALUE: 1.0
+          MAX: 100.0
+          STEP: 1.0
+          MIN: 1.0
+        TEMPERATURE:
+          VALUE: 1.0
+          MAX: 100.0
+          STEP: 1.0
+          MIN: 1.0
+      - PROMPT: Generate the caption in English
+        LANGUAGE_NAME: English
+        LANGUAGE_ZH_NAME: 英语
+        MAX_NEW_TOKENS:
+          VALUE: 1024
+          MAX: 2048
+          STEP: 128
+          MIN: 256
+        MIN_NEW_TOKENS:
+          VALUE: 16
+          MAX: 1024
+          STEP: 16
+          MIN: 0
+        NUM_BEAMS:
+          VALUE: 1
+          MAX: 12
+          STEP: 1
+          MIN: 1
+        REPETITION_PENALTY:
+          VALUE: 1.0
+          MAX: 100.0
+          STEP: 1.0
+          MIN: 1.0
+        TEMPERATURE:
+          VALUE: 1.0
+          MAX: 100.0
+          STEP: 1.0
+          MIN: 1.0
+
   - NAME: QWVL
     TYPE: caption
     MODEL_PATH: ms://qwen/Qwen-VL:v1.0.3
@@ -77,75 +151,14 @@ PROCESSORS:
           MAX: 100.0
           STEP: 1.0
           MIN: 1.0
-  -
-    NAME: QWVLQuantize
-    TYPE: caption
-    DEVICE: "gpu"
-    MEMORY: 7885
-    MODEL_PATH: ms://qwen/Qwen-VL:v1.0.3
-    PARAS:
-      - PROMPT: 用中文描述这张图片
-        LANGUAGE_NAME: Chinese
-        LANGUAGE_ZH_NAME: 中文
-        MAX_NEW_TOKENS:
-          VALUE: 1024
-          MAX: 2048
-          STEP: 128
-          MIN: 256
-        MIN_NEW_TOKENS:
-          VALUE: 16
-          MAX: 1024
-          STEP: 16
-          MIN: 0
-        NUM_BEAMS:
-          VALUE: 1
-          MAX: 12
-          STEP: 1
-          MIN: 1
-        REPETITION_PENALTY:
-          VALUE: 1.0
-          MAX: 100.0
-          STEP: 1.0
-          MIN: 1.0
-        TEMPERATURE:
-          VALUE: 1.0
-          MAX: 100.0
-          STEP: 1.0
-          MIN: 1.0
-      - PROMPT: Generate the caption in English
-        LANGUAGE_NAME: English
-        LANGUAGE_ZH_NAME: 英语
-        MAX_NEW_TOKENS:
-          VALUE: 1024
-          MAX: 2048
-          STEP: 128
-          MIN: 256
-        MIN_NEW_TOKENS:
-          VALUE: 16
-          MAX: 1024
-          STEP: 16
-          MIN: 0
-        NUM_BEAMS:
-          VALUE: 1
-          MAX: 12
-          STEP: 1
-          MIN: 1
-        REPETITION_PENALTY:
-          VALUE: 1.0
-          MAX: 100.0
-          STEP: 1.0
-          MIN: 1.0
-        TEMPERATURE:
-          VALUE: 1.0
-          MAX: 100.0
-          STEP: 1.0
-          MIN: 1.0
-  -
-    NAME: CenterCrop
+
+  # Simple processor
+  - NAME: CenterCrop
     TYPE: image
     DEVICE: "cpu"
     MEMORY: 10
     PARAS:
+      CAPTION_INTERACTIVE: False
       HEIGHT_RATIO:
         VALUE: 1
         MAX: 20
@@ -156,18 +169,309 @@ PROCESSORS:
         MAX: 20
         STEP: 1
         MIN: 1
-#  - NAME: PaddingCrop
-#    TYPE: image
-#    DEVICE: "cpu"
-#    MEMORY: 10
-#    PARAS:
-#      HEIGHT_RATIO:
-#        VALUE: 3
-#        MAX: 25
-#        STEP: 1
-#        MIN: 1
-#      WIDTH_RATIO:
-#        VALUE: 4
-#        MAX: 20
-#        STEP: 1
-#        MIN: 1
+  - NAME: ChangeSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      CAPTION_INTERACTIVE: False
+      SRC_IMAGE_INTERACTIVE: True
+      SRC_IMAGE_MASK_INTERACTIVE: True
+      TARGET_IMAGE_INTERACTIVE: True
+      PREVIEW_BTN_VISIBLE: False
+  - NAME: MaskEditSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      SRC_IMAGE_INTERACTIVE: True
+      SRC_IMAGE_TOOL: sketch
+      CAPTION_INTERACTIVE: False
+  - NAME: SourceMaskSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      SRC_IMAGE_INTERACTIVE: True
+      SRC_IMAGE_TOOL: sketch
+      CAPTION_INTERACTIVE: False
+  - NAME: MaskSwapEditSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      PREVIEW_BTN_VISIBLE: False
+      SRC_IMAGE_INTERACTIVE: True
+      SRC_IMAGE_TOOL: sketch
+      CAPTION_INTERACTIVE: False
+  - NAME: SwapMaskSwapEditSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      PREVIEW_BTN_VISIBLE: False
+      SRC_IMAGE_INTERACTIVE: True
+      SRC_IMAGE_TOOL: sketch
+      CAPTION_INTERACTIVE: False
+  - NAME: SwapSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      PREVIEW_BTN_VISIBLE: True
+      SRC_IMAGE_INTERACTIVE: True
+      TARGET_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: DefaultMaskSample
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      PREVIEW_BTN_VISIBLE: True
+      SRC_IMAGE_INTERACTIVE: False
+      TARGET_IMAGE_INTERACTIVE: False
+      CAPTION_INTERACTIVE: False
+  #  - NAME: PaddingCrop
+  #    TYPE: image
+  #    DEVICE: "cpu"
+  #    MEMORY: 10
+  #    PARAS:
+  #      HEIGHT_RATIO:
+  #        VALUE: 3
+  #        MAX: 25
+  #        STEP: 1
+  #        MIN: 1
+  #      WIDTH_RATIO:
+  #        VALUE: 4
+  #        MAX: 20
+  #        STEP: 1
+  #        MIN: 1
+  # Annotator processor
+  - NAME: CannyExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "CannyAnnotator"
+      LOW_THRESHOLD: 100
+      HIGH_THRESHOLD: 200
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: ColorExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "ColorAnnotator"
+      RATIO: 64
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: InfoDrawContourExtractor
+    TYPE: image
+    DEVICE: "gpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "InfoDrawContourAnnotator"
+      INPUT_NC: 3
+      OUTPUT_NC: 1
+      N_RESIDUAL_BLOCKS: 3
+      SIGMOID: True
+      PRETRAINED_MODEL: "ms://iic/scepter_annotator@annotator/ckpts/informative_drawing_contour_style.pth"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: DegradationExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "DegradationAnnotator"
+      RANDOM_DEGRADATION: True
+      PARAMS:
+        gaussian_noise: { }
+        resize: { 'scale': [ 0.4, 0.8 ] }
+        jpeg: { 'jpeg_level': [ 25, 75 ] }
+        gaussian_blur: { 'kernel_size': [ 7, 9, 11, 13, 15 ], 'sigma': [ 0.9, 1.8 ] }
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: MidasExtractor
+    TYPE: image
+    DEVICE: "gpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "MidasDetector"
+      PRETRAINED_MODEL: "ms://iic/scepter_annotator@annotator/ckpts/dpt_hybrid-midas-501f0c75.pt"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: DoodleExtractor
+    TYPE: image
+    DEVICE: "gpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "DoodleAnnotator"
+      PROCESSOR_TYPE: "pidinet_sketch"
+      PROCESSOR_CFG:
+        - NAME: "PiDiAnnotator"
+          PRETRAINED_MODEL: "ms://iic/scepter_annotator@annotator/ckpts/table5_pidinet.pth"
+        - NAME: "SketchAnnotator"
+          PRETRAINED_MODEL: "ms://iic/scepter_annotator@annotator/ckpts/sketch_simplification_gan.pth"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: GrayExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "GrayAnnotator"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: InpaintingExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "InpaintingAnnotator"
+      RETURN_MASK: False
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+
+  - NAME: InpaintingSourceExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: OpenposeExtractor
+    TYPE: image
+    DEVICE: "gpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "OpenposeAnnotator"
+      BODY_MODEL_PATH: "ms://iic/scepter_annotator@annotator/ckpts/body_pose_model.pth"
+      HAND_MODEL_PATH: "ms://iic/scepter_annotator@annotator/ckpts/hand_pose_model.pth"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: OutpaintingExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "OutpaintingAnnotator"
+      RETURN_MASK: False
+      KEEP_PADDING_RATIO: 1
+      RANDOM_CFG:
+        DIRECTION_RANGE: [ 'left', 'right', 'up', 'down' ]
+        RATIO_RANGE: [ 0.1, 0.7 ]
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+      USE_MASK_VISIBLE: True
+      PREVIEW_BTN_VISIBLE: False
+  - NAME: OutpaintingResize
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "OutpaintingResize"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+      PREVIEW_BTN_VISIBLE: False
+  - NAME: InfoDrawAnimeAnnotator
+    TYPE: image
+    DEVICE: "gpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "InfoDrawAnimeAnnotator"
+      INPUT_NC: 3
+      OUTPUT_NC: 1
+      N_RESIDUAL_BLOCKS: 3
+      SIGMOID: True
+      PRETRAINED_MODEL: "ms://iic/scepter_annotator@annotator/ckpts/informative_drawing_anime_style.pth"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: ESAMExtractor
+    TYPE: image
+    DEVICE: "gpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "ESAMAnnotator"
+      PRETRAINED_MODEL: "ms://iic/scepter_annotator@annotator/ckpts/efficient_sam_vits.pt"
+      SAVE_MODE: 'P'
+      GRID_SIZE: 32
+      USE_DOMINANT_COLOR: True
+      RETURN_MASK: False
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+  - NAME: InvertExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "InvertAnnotator"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+
+  - NAME: LamaExtractor
+    TYPE: image
+    DEVICE: "cpu"
+    MEMORY: 10
+    MODEL:
+      NAME: "LamaAnnotator"
+      PRETRAINED_MODEL: "ms:///iic/cv_fft_inpainting_lama/"
+    PARAS:
+      SRC_IMAGE_TOOL: sketch
+      SRC_IMAGE_INTERACTIVE: True
+      CAPTION_INTERACTIVE: False
+
+VIDEO_PROCESSORS:
+  - NAME: CogVLM2Llama3Caption
+    TYPE: caption
+    MODEL_PATH: ms://ZhipuAI/cogvlm2-llama3-caption
+    DEVICE: "gpu"
+    MEMORY: 20000
+    PROMPT: Please describe this video in detail.
+    TEMPERATURE: 0.1
+    MAX_NEW_TOKENS: 2048
+    PAD_TOKEN_ID: 128002
+    TOP_K: 1
+    TOP_P: 0.1
+
+TRANSLATION_PROCESSORS:
+  - NAME: OpusMtZhEn
+    TYPE: caption
+    MODEL_PATH: ms://cubeai/trans-opus-mt-zh-en
+    DEVICE: "gpu"
+    MEMORY: 5000
+  - NAME: OpusMtEnZh
+    TYPE: caption
+    MODEL_PATH: ms://cubeai/trans-opus-mt-en-zh
+    DEVICE: "gpu"
+    MEMORY: 5000

scepter/methods/studio/scepter_ui.yaml

@@ -87,3 +87,7 @@ INTERFACE:
     NAME_EN: Inference
     IFID: inference
     CONFIG: scepter/methods/studio/inference/inference.yaml
+  - NAME: 对话式编辑
+    NAME_EN: ChatBot
+    IFID: chatbot
+    CONFIG: scepter/methods/studio/chatbot/chatbot.yaml

scepter/methods/studio/self_train/dit/cogvideox_2b_pro.yaml

@@ -0,0 +1,315 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+META:
+  VERSION: 'COGVIDEOX_2B'
+  DESCRIPTION: "cogvideox 2b"
+  IS_DEFAULT: False
+  IS_SHARE: True
+  INFERENCE_PARAS:
+    INFERENCE_BATCH_SIZE: 1
+    INFERENCE_PREFIX: ""
+    DEFAULT_SAMPLER: "ddim"
+    DEFAULT_SAMPLE_STEPS: 50
+    INFERENCE_N_PROMPT: ""
+    RESOLUTION: [ 480, 720 ]
+  PARAS:
+    - TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [ 480, 720 ]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: False
+      TUNER: FULL
+    - TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [ 480, 720 ]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: True
+      TUNER: LORA
+  #
+  TUNERS:
+    LORA:
+      - NAME: SwiftLoRA
+        R: 64
+        LORA_ALPHA: 64
+        LORA_DROPOUT: 0.0
+        BIAS: "none"
+        TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox_2b_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  FPS: 8
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 1.15258426
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 3.0
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@transformer/diffusion_pytorch_model.safetensors
+      NUM_ATTENTION_HEADS: 30
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      NUM_LAYERS: 30
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 90
+      SAMPLE_HEIGHT: 60
+      SAMPLE_FRAMES: 49
+      PATCH_SIZE: 2
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 226
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: False
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+      GRADIENT_CHECKPOINTING: False
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@vae/diffusion_pytorch_model.safetensors
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: False
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: False
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 226
+      CLEAN:
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 49
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDatasetOTF
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    PROMPT_PREFIX: ''
+    DELIMITER: '#;#'
+    FIELDS: [ 'video_path', 'width', 'height', 'prompt' ]
+    PATH_PREFIX:
+    DATA_FILE:
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'video', 'video_latent', "prompt" ]
+        META_KEYS: [ ]
+    MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-2b@vae/diffusion_pytorch_model.safetensors
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "A girl riding a bike.", "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical atmosphere of this unique musical performance." ]
+    IMAGE_SIZE: [ 480, 720 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+#    USE_NUM: 8
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/studio/self_train/dit/cogvideox_5b_pro.yaml

@@ -0,0 +1,317 @@
+ENV:
+  BACKEND: nccl
+  SEED: 42
+  TENSOR_PARALLEL_SIZE: 1
+  PIPELINE_PARALLEL_SIZE: 1
+  SYS_ENVS:
+    TORCH_CUDNN_V8_API_ENABLED: '1'
+    TOKENIZERS_PARALLELISM: 'false'
+    TF_CPP_MIN_LOG_LEVEL: '3'
+    PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
+META:
+  VERSION: 'COGVIDEOX_5B'
+  DESCRIPTION: "cogvideox 5b"
+  IS_DEFAULT: False
+  IS_SHARE: True
+  INFERENCE_PARAS:
+    INFERENCE_BATCH_SIZE: 1
+    INFERENCE_PREFIX: ""
+    DEFAULT_SAMPLER: "ddim"
+    DEFAULT_SAMPLE_STEPS: 50
+    INFERENCE_N_PROMPT: ""
+    RESOLUTION: [ 480, 720 ]
+  PARAS:
+    - TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [ 480, 720 ]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: False
+      TUNER: FULL
+    - TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [ 480, 720 ]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: True
+      TUNER: LORA
+  #
+  TUNERS:
+    LORA:
+      - NAME: SwiftLoRA
+        R: 64
+        LORA_ALPHA: 64
+        LORA_DROPOUT: 0.0
+        BIAS: "none"
+        TARGET_MODULES: "model.*(.to_k|.to_q|.to_v|.to_out.0)$"
+#
+SOLVER:
+  NAME: LatentDiffusionVideoSolver
+  MAX_STEPS: 2000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_cogvideox_5b_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 4
+  FPS: 8
+  SHARDING_STRATEGY: full_shard
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.model']
+  SAVE_MODULES: [ 'model', 'cond_stage_model.model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  TUNER:
+  #
+  MODEL:
+    NAME: LatentDiffusionCogVideoX
+    PRETRAINED_MODEL:
+    PARAMETERIZATION: v
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA: 3.0
+    ZERO_TERMINAL_SNR: True
+    SCALE_FACTOR_SPATIAL: 8
+    SCALE_FACTOR_TEMPORAL: 4
+    SCALING_FACTOR_IMAGE: 0.7  # 5b diff
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: BaseDiffusion
+      PREDICTION_TYPE: v
+      NOISE_SCHEDULER:
+        NAME: ScaledLinearScheduler
+        BETA_MIN: 0.00085
+        BETA_MAX: 0.012
+        SNR_SHIFT_SCALE: 1.0  # 5b diff
+        RESCALE_BETAS_ZERO_SNR: True
+      DIFFUSION_SAMPLERS:
+        NAME: DDIMSampler
+        DISCRETIZATION_TYPE: trailing
+        ETA: 0.0
+    #
+    DIFFUSION_MODEL:
+      NAME: CogVideoXTransformer3DModel
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: # 5b diff
+        - ms://AI-ModelScope/CogVideoX-5b@transformer/diffusion_pytorch_model-00001-of-00002.safetensors
+        - ms://AI-ModelScope/CogVideoX-5b@transformer/diffusion_pytorch_model-00002-of-00002.safetensors
+      NUM_ATTENTION_HEADS: 48  # 5b diff
+      ATTENTION_HEAD_DIM: 64
+      IN_CHANNELS: 16
+      OUT_CHANNELS: 16
+      FLIP_SIN_TO_COS: True
+      FREQ_SHIFT: 0
+      TIME_EMBED_DIM: 512
+      TEXT_EMBED_DIM: 4096
+      NUM_LAYERS: 42  # 5b diff
+      DROPOUT: 0.0
+      ATTENTION_BIAS: True
+      SAMPLE_WIDTH: 90
+      SAMPLE_HEIGHT: 60
+      SAMPLE_FRAMES: 49
+      PATCH_SIZE: 2
+      TEMPORAL_COMPRESSION_RATIO: 4
+      MAX_TEXT_SEQ_LENGTH: 226
+      ACTIVATION_FN: "gelu-approximate"
+      TIMESTEP_ACTIVATION_FN: "silu"
+      NORM_ELEMENTWISE_AFFINE: True
+      NORM_EPS: 1e-5
+      SPATIAL_INTERPOLATION_SCALE: 1.875
+      TEMPORAL_INTERPOLATION_SCALE: 1.0
+      USE_ROTARY_POSITIONAL_EMBEDDINGS: True # 5b diff
+      USE_LEARNED_POSITIONAL_EMBEDDINGS: False
+      GRADIENT_CHECKPOINTING: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-5b@vae/diffusion_pytorch_model.safetensors # 5b diff
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/t5-v1_1-xxl
+      TOKENIZER_PATH: ms://AI-ModelScope/t5-v1_1-xxl
+      LENGTH: 226
+      CLEAN:
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 50
+    SEED: 42
+    GUIDE_SCALE: 6.0
+    GUIDE_RESCALE: 0.0
+    NUM_FRAMES: 49
+  #
+  OPTIMIZER:
+    NAME: Adam
+    LEARNING_RATE: 1e-3
+    BETAS: [ 0.9, 0.95 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 0.0
+    AMSGRAD: False
+  #
+#  LR_SCHEDULER:
+#    NAME: StepAnnealingLR
+#    WARMUP_STEPS: 200
+#    TOTAL_STEPS: 2000
+#    DECAY_MODE: 'cosine'
+  #
+  TRAIN_DATA:
+    NAME: VideoGenDatasetOTF
+    MODE: train
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    PROMPT_PREFIX: ''
+    DELIMITER: '#;#'
+    FIELDS: [ 'video_path', 'width', 'height', 'prompt' ]
+    PATH_PREFIX:
+    DATA_FILE:
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'video', 'video_latent', "prompt" ]
+        META_KEYS: [ ]
+    MODEL:
+      NAME: AutoencoderKLCogVideoX
+      DTYPE: bfloat16
+      PRETRAINED_MODEL: ms://AI-ModelScope/CogVideoX-5b@vae/diffusion_pytorch_model.safetensors
+      SAMPLE_HEIGHT: 480
+      SAMPLE_WIDTH: 720
+      USE_QUANT_CONV: False
+      USE_POST_QUANT_CONV: False
+      USE_SLICING: True
+      USE_TILING: True
+      GRADIENT_CHECKPOINTING: True
+      ENCODER:
+        NAME: CogVideoXEncoder3D
+        IN_CHANNELS: 3
+        OUT_CHANNELS: 16
+        UP_BLOCK_TYPES: [ "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D", "CogVideoXDownBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+      DECODER:
+        NAME: CogVideoXDecoder3D
+        IN_CHANNELS: 16
+        OUT_CHANNELS: 3
+        UP_BLOCK_TYPES: [ "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D", "CogVideoXUpBlock3D" ]
+        BLOCK_OUT_CHANNELS: [ 128, 256, 256, 512 ]
+        LAYERS_PER_BLOCK: 3
+        ACT_FN: "silu"
+        NORM_EPS: 1e-6
+        NORM_NUM_GROUPS: 32
+        DROPOUT: 0.0
+        PAD_MODE: "first"
+        TEMPORAL_COMPRESSION_RATIO: 4
+        GRADIENT_CHECKPOINTING: True
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "A girl riding a bike.", "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical atmosphere of this unique musical performance." ]
+    IMAGE_SIZE: [ 480, 720 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    #    USE_NUM: 8
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+      PRIORITY: 20
+    - NAME: CheckpointHook
+      INTERVAL: 1000
+      PRIORITY: 40
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/studio/self_train/dit/flux1.0_dv_pro.yaml

@@ -0,0 +1,285 @@
+ENV:
+  BACKEND: nccl
+META:
+  VERSION: 'FLUX1.0_DEV'
+  DESCRIPTION: "flux 1.0 dev"
+  IS_DEFAULT: True
+  IS_SHARE: True
+  INFERENCE_PARAS:
+    INFERENCE_BATCH_SIZE: 1
+    INFERENCE_PREFIX: ""
+    DEFAULT_SAMPLER: "flow_euler"
+    DEFAULT_SAMPLE_STEPS: 50
+    INFERENCE_N_PROMPT: ""
+    RESOLUTION: [1024, 1024]
+  PARAS:
+    -
+      TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: False
+      TUNER: FULL
+    -
+      TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: True
+      TUNER: LORA
+  #
+  TUNERS:
+    LORA:
+      -
+        NAME: SwiftLoRA
+        R: 4
+        LORA_ALPHA: 4
+        LORA_DROPOUT: 0.0
+        BIAS: "none"
+        TARGET_MODULES: "(model.double_blocks.*(.qkv|.proj|.img_mod.lin|.txt_mod.lin))|(model.single_blocks.*(.linear1|.linear2|.modulation.lin))$"
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  MAX_STEPS: 100000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_flux_dev_1024_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 16
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.t5_model' ] #
+  SAVE_MODULES: [ 'model']
+  TRAIN_MODULES: ['model']
+
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+
+  FREEZE:
+
+  TUNER:
+
+  MODEL:
+    NAME: LatentDiffusionFlux
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: DiffusionFluxRF
+      PREDICTION_TYPE: raw
+      NOISE_SCHEDULER:
+        NAME: FlowMatchSigmaScheduler
+        WEIGHTING_SCHEME: logit_normal
+        SHIFT: 3.0
+        LOGIT_MEAN: 0.0
+        LOGIT_STD: 1.0
+        MODE_SCALE: 1.29
+      SAMPLER_SCHEDULER:
+        NAME: FlowMatchFluxShiftScheduler
+        SHIFT: False
+        SIGMOID_SCALE: 1
+        BASE_SHIFT: 0.5
+        MAX_SHIFT: 1.15
+
+    DIFFUSION_MODEL:
+      NAME: Flux
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@flux1-dev.safetensors
+      IN_CHANNELS: 64
+      HIDDEN_SIZE: 3072
+      NUM_HEADS: 24
+      AXES_DIM: [ 16, 56, 56 ]
+      THETA: 10000
+      VEC_IN_DIM: 768
+      GUIDANCE_EMBED: False
+      CONTEXT_IN_DIM: 4096
+      MLP_RATIO: 4.0
+      QKV_BIAS: True
+      DEPTH: 19
+      DEPTH_SINGLE_BLOCKS: 38
+      USE_GRAD_CHECKPOINT: True
+
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLFlux
+      EMBED_DIM: 16
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-dev@ae.safetensors
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 8
+      USE_CONV: False
+      SCALE_FACTOR: 0.3611
+      SHIFT_FACTOR: 0.1159
+
+      ENCODER:
+        NAME: Encoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+
+      DECODER:
+        NAME: Decoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+
+    COND_STAGE_MODEL:
+      NAME: T5PlusClipFluxEmbedder
+      T5_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: T5EncoderModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder_2/
+        HF_TOKENIZER_CLS: T5Tokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer_2/
+        MAX_LENGTH: 512
+        OUTPUT_KEY: last_hidden_state
+        D_TYPE: bfloat16
+        BATCH_INFER: False
+        CLEAN: whitespace
+      CLIP_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: CLIPTextModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-dev@text_encoder/
+        HF_TOKENIZER_CLS: CLIPTokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-dev@tokenizer/
+        MAX_LENGTH: 77
+        OUTPUT_KEY: pooler_output
+        D_TYPE: bfloat16
+        BATCH_INFER: True
+        CLEAN: whitespace
+
+  SAMPLE_ARGS:
+    SAMPLE_STEPS: 50
+    SAMPLER: flow_euler
+    SEED: 2024
+    IMAGE_SIZE: [ 1024, 1024 ]
+    SHIFT: True
+    GUIDE_SCALE: 3.5
+
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 4e-4
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a cat holds a blackboard that writes \"hello world\"", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 2
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+    - NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/studio/self_train/dit/flux1.0_schnell_pro.yaml

@@ -0,0 +1,284 @@
+ENV:
+  BACKEND: nccl
+META:
+  VERSION: 'FLUX1.0_SCHNELL'
+  DESCRIPTION: "flux 1.0 schnell"
+  IS_DEFAULT: False
+  IS_SHARE: True
+  INFERENCE_PARAS:
+    INFERENCE_BATCH_SIZE: 1
+    INFERENCE_PREFIX: ""
+    DEFAULT_SAMPLER: "flow_euler"
+    DEFAULT_SAMPLE_STEPS: 4
+    INFERENCE_N_PROMPT: ""
+    RESOLUTION: [1024, 1024]
+  PARAS:
+    -
+      TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: False
+      TUNER: FULL
+    -
+      TRAIN_BATCH_SIZE: 1
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 89000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 4e-4
+      IS_DEFAULT: True
+      TUNER: LORA
+  #
+  TUNERS:
+    LORA:
+      -
+        NAME: SwiftLoRA
+        R: 4
+        LORA_ALPHA: 4
+        LORA_DROPOUT: 0.0
+        BIAS: "none"
+        TARGET_MODULES: "(model.double_blocks.*(.qkv|.proj|.img_mod.lin|.txt_mod.lin))|(model.single_blocks.*(.linear1|.linear2|.modulation.lin))$"
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  MAX_STEPS: 100000
+  USE_AMP: True
+  DTYPE: bfloat16
+  USE_FAIRSCALE: False
+  USE_FSDP: True
+  LOAD_MODEL_ONLY: False
+  RESUME_FROM:
+  WORK_DIR: ./cache/save_data/dit_flux_schnell_1024_lora
+  LOG_FILE: std_log.txt
+  EVAL_INTERVAL: 100
+  LOG_TRAIN_NUM: 16
+  ENABLE_GRADSCALER: False
+  USE_SCALER: False
+  FSDP_REDUCE_DTYPE: float32
+  FSDP_BUFFER_DTYPE: float32
+  FSDP_SHARD_MODULES: [ 'model', 'cond_stage_model.t5_model' ]
+  SAVE_MODULES: [ 'model']
+  TRAIN_MODULES: ['model']
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+
+  FREEZE:
+
+  TUNER:
+
+  MODEL:
+    NAME: LatentDiffusionFlux
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    DEFAULT_N_PROMPT:
+    USE_EMA: False
+    EVAL_EMA: False
+    DIFFUSION:
+      NAME: DiffusionFluxRF
+      PREDICTION_TYPE: raw
+      NOISE_SCHEDULER:
+        NAME: FlowMatchSigmaScheduler
+        WEIGHTING_SCHEME: logit_normal
+        SHIFT: 3.0
+        LOGIT_MEAN: 0.0
+        LOGIT_STD: 1.0
+        MODE_SCALE: 1.29
+      SAMPLER_SCHEDULER:
+        NAME: FlowMatchFluxShiftScheduler
+        SHIFT: False
+        SIGMOID_SCALE: 1
+        BASE_SHIFT: 0.5
+        MAX_SHIFT: 1.15
+
+    DIFFUSION_MODEL:
+      NAME: Flux
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-schnell@flux1-schnell.safetensors
+      IN_CHANNELS: 64
+      HIDDEN_SIZE: 3072
+      NUM_HEADS: 24
+      AXES_DIM: [ 16, 56, 56 ]
+      THETA: 10000
+      VEC_IN_DIM: 768
+      GUIDANCE_EMBED: False
+      CONTEXT_IN_DIM: 4096
+      MLP_RATIO: 4.0
+      QKV_BIAS: True
+      DEPTH: 19
+      DEPTH_SINGLE_BLOCKS: 38
+      USE_GRAD_CHECKPOINT: True
+
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKLFlux
+      EMBED_DIM: 16
+      PRETRAINED_MODEL: ms://AI-ModelScope/FLUX.1-schnell@ae.safetensors
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 8
+      USE_CONV: False
+      SCALE_FACTOR: 0.3611
+      SHIFT_FACTOR: 0.1159
+
+      ENCODER:
+        NAME: Encoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+
+      DECODER:
+        NAME: Decoder
+        USE_CHECKPOINT: True
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+
+    COND_STAGE_MODEL:
+      NAME: T5PlusClipFluxEmbedder
+      T5_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: T5EncoderModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-schnell@text_encoder_2/
+        HF_TOKENIZER_CLS: T5Tokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-schnell@tokenizer_2/
+        MAX_LENGTH: 256
+        OUTPUT_KEY: last_hidden_state
+        D_TYPE: bfloat16
+        BATCH_INFER: False
+        CLEAN: whitespace
+      CLIP_MODEL:
+        NAME: HFEmbedder
+        HF_MODEL_CLS: CLIPTextModel
+        MODEL_PATH: ms://AI-ModelScope/FLUX.1-schnell@text_encoder/
+        HF_TOKENIZER_CLS: CLIPTokenizer
+        TOKENIZER_PATH: ms://AI-ModelScope/FLUX.1-schnell@tokenizer/
+        MAX_LENGTH: 77
+        OUTPUT_KEY: pooler_output
+        D_TYPE: bfloat16
+        BATCH_INFER: True
+        CLEAN: whitespace
+
+  SAMPLE_ARGS:
+    SAMPLE_STEPS: 4
+    SAMPLER: flow_euler
+    SEED: 2024
+    IMAGE_SIZE: [ 1024, 1024 ]
+    GUIDE_SCALE: 3.5
+
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 4e-4
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a cat holds a blackboard that writes \"hello world\"", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 2
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+
+  TRAIN_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      PRIORITY: 0
+    - NAME: BackwardHook
+      PRIORITY: 10
+    - NAME: LogHook
+      LOG_INTERVAL: 10
+    - NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  EVAL_HOOKS:
+    - NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/studio/self_train/dit/pixart_alpha_pro.yaml

@@ -0,0 +1,264 @@
+ENV:
+  BACKEND: nccl
+META:
+  VERSION: 'PIXART_ALPHA'
+  DESCRIPTION: "PIXART ALPHA"
+  IS_DEFAULT: False
+  IS_SHARE: True
+  INFERENCE_PARAS:
+    INFERENCE_BATCH_SIZE: 1
+    INFERENCE_PREFIX: ""
+    DEFAULT_SAMPLER: "ddim"
+    DEFAULT_SAMPLE_STEPS: 20
+    INFERENCE_N_PROMPT: ""
+    RESOLUTION: [1024, 1024]
+  PARAS:
+    -
+      TRAIN_BATCH_SIZE: 2
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 29000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 0.0001
+      IS_DEFAULT: False
+      TUNER: FULL
+    -
+      TRAIN_BATCH_SIZE: 2
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 29000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 0.0001
+      IS_DEFAULT: True
+      TUNER: LORA
+  #
+  TUNERS:
+    LORA:
+      -
+        NAME: SwiftLoRA
+        R: 256
+        LORA_ALPHA: 256
+        LORA_DROPOUT: 0.0
+        BIAS: "none"
+        TARGET_MODULES: "model.*(.q|.k|.v|.o|mlp.fc1|mlp.fc2)$"
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 1000
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: -1
+  RESCALE_LR: False
+  #
+  WORK_DIR: cache/scepter_ui/self_train/dit/pixart_alpha_pro
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    NAME: "ModelscopeFs"
+    TEMP_DIR: "./cache/cache_data"
+  #
+  FREEZE:
+  #
+  TUNER:
+  #
+  MODEL:
+    NAME: LatentDiffusionPixart
+    PARAMETERIZATION: eps
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 0.18215
+    SIZE_FACTOR: 8
+    DECODER_BIAS: 0.5
+    DEFAULT_N_PROMPT:
+    SCHEDULE_ARGS:
+      "NAME": "linear"
+      "BETA_MIN": 0.0001
+      "BETA_MAX": 0.02
+    USE_EMA: False
+    LOAD_REFINER: False
+    #
+    DIFFUSION_MODEL:
+      NAME: PixArt
+      PRETRAINED_MODEL: ms://AI-ModelScope/PixArt-alpha@PixArt-XL-2-1024-MS.pth
+      INPUT_SIZE: 128
+      PATCH_SIZE: 2
+      IN_CHANNELS: 4
+      HIDDEN_SIZE: 1152
+      DEPTH: 28
+      NUM_HEADS: 16
+      MLP_RATIO: 4.0
+      CLASS_DROPOUT_PROB: 0.1
+      PRED_SIGMA: True
+      DROP_PATH: 0.0
+      WINDOW_DIZE: 0
+      USE_REL_POS: False
+      CAPTION_CHANNELS: 4096
+      LEWEI_SCALE: 2
+      MODEL_MAX_LENGTH: 120
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-2-base@512-base-ema.safetensors
+      EMBED_DIM: 4
+      IGNORE_KEYS: [ ]
+      BATCH_SIZE: 1
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 4
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: T5EmbedderHF
+      PRETRAINED_MODEL: ms://AI-ModelScope/PixArt-alpha@t5-v1_1-xxl/
+      TOKENIZER_PATH: ms://AI-ModelScope/PixArt-alpha@t5-v1_1-xxl/
+      LENGTH: 120
+      CLEAN: heavy
+      USE_GRAD: False
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: ddim
+    SAMPLE_STEPS: 20
+    SEED: 2024
+    GUIDE_SCALE: 4.5
+    GUIDE_RESCALE: 0.5
+    DISCRETIZATION: trailing
+    RUN_TRAIN_N: False
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 0.0001
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a boy wearing a jacket", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 0
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 10
+      SHOW_GPU_MEM: True
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/studio/self_train/dit/sd3_pro.yaml

@@ -0,0 +1,284 @@
+ENV:
+  BACKEND: nccl
+META:
+  VERSION: 'SD3'
+  DESCRIPTION: "SD3 ALPHA"
+  IS_DEFAULT: False
+  IS_SHARE: True
+  INFERENCE_PARAS:
+    INFERENCE_BATCH_SIZE: 1
+    INFERENCE_PREFIX: ""
+    DEFAULT_SAMPLER: "euler"
+    DEFAULT_SAMPLE_STEPS: 28
+    INFERENCE_N_PROMPT: ""
+    RESOLUTION: [1024, 1024]
+  PARAS:
+    -
+      TRAIN_BATCH_SIZE: 2
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 29000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 1e-5
+      IS_DEFAULT: False
+      TUNER: FULL
+    -
+      TRAIN_BATCH_SIZE: 2
+      TRAIN_PREFIX: ""
+      TRAIN_N_PROMPT: ""
+      RESOLUTION: [1024, 1024]
+      MEMORY: 29000
+      EPOCHS: 50
+      SAVE_INTERVAL: 25
+      EPSEC: 0.818
+      LEARNING_RATE: 5e-5
+      IS_DEFAULT: True
+      TUNER: LORA
+  #
+  TUNERS:
+    LORA:
+      -
+        NAME: SwiftLoRA
+        R: 128
+        LORA_ALPHA: 128
+        LORA_DROPOUT: 0.0
+        BIAS: "none"
+        TARGET_MODULES: "model.*(.attn.qkv|.attn.proj|mlp.fc1|mlp.fc2)$"
+#
+SOLVER:
+  NAME: LatentDiffusionSolver
+  RESUME_FROM:
+  LOAD_MODEL_ONLY: True
+  USE_FSDP: False
+  SHARDING_STRATEGY:
+  USE_AMP: True
+  DTYPE: float16
+  CHANNELS_LAST: True
+  MAX_STEPS: 1000
+  MAX_EPOCHS: -1
+  NUM_FOLDS: 1
+  ACCU_STEP: 1
+  EVAL_INTERVAL: -1
+  RESCALE_LR: False
+  #
+  WORK_DIR: cache/scepter_ui/self_train/dit/sd3
+  LOG_FILE: std_log.txt
+  #
+  FILE_SYSTEM:
+    - NAME: "ModelscopeFs"
+      TEMP_DIR: "./cache/cache_data"
+  #
+  FREEZE:
+  #
+  TUNER:
+  #
+  MODEL:
+    NAME: LatentDiffusionSD3
+    PARAMETERIZATION: rf
+    TIMESTEPS: 1000
+    MIN_SNR_GAMMA:
+    ZERO_TERMINAL_SNR: False
+    PRETRAINED_MODEL:
+    IGNORE_KEYS: [ ]
+    SCALE_FACTOR: 1.5305
+    SHIFT_FACTOR: 0.0609
+    DEFAULT_N_PROMPT:
+    SCHEDULE_ARGS:
+      "NAME": "shifted"
+      "SHIFT": 3
+    USE_EMA: False
+    T_WEIGHT: uniform
+    #
+    DIFFUSION_MODEL:
+      NAME: MMDiT
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+      IGNORE_KEYS: '^first_stage_model.'
+      IN_CHANNELS: 16
+      PATCH_SIZE: 2
+      OUT_CHANNELS: 16
+      DEPTH: 24
+      INPUT_SIZE:
+      ADM_IN_CHANNELS: 2048
+      CONTEXT_EMBEDDER_CONFIG: { 'target': 'torch.nn.Linear', 'params': { 'in_features': 4096, 'out_features': 1536 } }
+      NUM_PATCHES: 36864
+      POS_EMBED_MAX_SIZE: 192
+      POS_EMBED_SCALING_FACTOR:
+      USE_CHECKPOINT: True
+    #
+    FIRST_STAGE_MODEL:
+      NAME: AutoencoderKL
+      PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium@sd3_medium.safetensors
+      EMBED_DIM: 16
+      IGNORE_KEYS: '^model.diffusion_model.'
+      BATCH_SIZE: 1
+      USE_CONV: False
+      #
+      ENCODER:
+        NAME: Encoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DOUBLE_Z: True
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+      #
+      DECODER:
+        NAME: Decoder
+        CH: 128
+        OUT_CH: 3
+        NUM_RES_BLOCKS: 2
+        IN_CHANNELS: 3
+        ATTN_RESOLUTIONS: [ ]
+        CH_MULT: [ 1, 2, 4, 4 ]
+        Z_CHANNELS: 16
+        DROPOUT: 0.0
+        RESAMP_WITH_CONV: True
+        GIVE_PRE_END: False
+        TANH_OUT: False
+    #
+    COND_STAGE_MODEL:
+      NAME: SD3TextEmbedder
+      P_ZERO: 0.0
+      CLIP_L:
+        NAME: FrozenCLIPEmbedder2
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer
+        MAX_LENGTH: 77
+        FREEZE: True
+        LAYER: penultimate
+        RETURN_POOLED: True
+        USE_FINAL_LAYER_NORM: False
+        IS_TRAINABLE: False
+      CLIP_G:
+        NAME: FrozenCLIPEmbedder2
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_2
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_2
+        MAX_LENGTH: 77
+        FREEZE: True
+        LAYER: penultimate
+        RETURN_POOLED: True
+        USE_FINAL_LAYER_NORM: False
+        IS_TRAINABLE: False
+      T5_XXL:
+        NAME: T5EmbedderHF
+        PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@text_encoder_3
+        TOKENIZER_PATH: ms://AI-ModelScope/stable-diffusion-3-medium-diffusers@tokenizer_3
+        LENGTH: 256
+        CLEAN: whitespace
+        USE_GRAD: False
+        T5_DTYPE: float16
+    #
+    LOSS:
+      NAME: ReconstructLoss
+      LOSS_TYPE: l2
+  #
+  SAMPLE_ARGS:
+    SAMPLER: euler
+    SAMPLE_STEPS: 28
+    SEED: 1749023094
+    GUIDE_SCALE: 5.0
+    GUIDE_RESCALE: 0.0
+    DISCRETIZATION: trailing
+    RUN_TRAIN_N: False
+  #
+  OPTIMIZER:
+    NAME: AdamW
+    LEARNING_RATE: 5e-5
+    BETAS: [ 0.9, 0.999 ]
+    EPS: 1e-8
+    WEIGHT_DECAY: 1e-2
+    AMSGRAD: False
+  #
+  TRAIN_DATA:
+    NAME: ImageTextPairMSDataset
+    MODE: train
+    MS_DATASET_NAME: style_custom_dataset
+    MS_DATASET_NAMESPACE: damo
+    MS_DATASET_SUBNAME: 3D
+    PROMPT_PREFIX: ""
+    MS_DATASET_SPLIT: train
+    MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' }
+    REPLACE_STYLE: False
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    SAMPLER:
+      NAME: LoopSampler
+    TRANSFORMS:
+      - NAME: LoadImageFromFile
+        RGB_ORDER: RGB
+        BACKEND: pillow
+      - NAME: FlexibleResize
+        INTERPOLATION: bilinear
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: FlexibleCenterCrop
+        SIZE: [ 1024, 1024 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: ImageToTensor
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'img' ]
+        BACKEND: pillow
+      - NAME: Normalize
+        MEAN: [ 0.5,  0.5,  0.5 ]
+        STD: [ 0.5,  0.5,  0.5 ]
+        INPUT_KEY: [ 'img' ]
+        OUTPUT_KEY: [ 'image' ]
+        BACKEND: torchvision
+      - NAME: Select
+        KEYS: [ 'image', 'prompt' ]
+        META_KEYS: [ 'data_key' ]
+  #
+  EVAL_DATA:
+    NAME: Text2ImageDataset
+    MODE: eval
+    PROMPT_FILE:
+    PROMPT_DATA: [ "a cat holds a blackboard that writes \"hello world\"", "a dog running on the lawn" ]
+    IMAGE_SIZE: [ 1024, 1024 ]
+    FIELDS: [ "prompt" ]
+    DELIMITER: '#;#'
+    PROMPT_PREFIX: ''
+    PIN_MEMORY: True
+    BATCH_SIZE: 1
+    NUM_WORKERS: 4
+    TRANSFORMS:
+      - NAME: Select
+        KEYS: [ 'index', 'prompt' ]
+        META_KEYS: [ 'image_size' ]
+  #
+  TRAIN_HOOKS:
+    -
+      NAME: BackwardHook
+      PRIORITY: 1
+    -
+      NAME: LogHook
+      LOG_INTERVAL: 10
+      SHOW_GPU_MEM: True
+    -
+      NAME: TensorboardLogHook
+    -
+      NAME: CheckpointHook
+      INTERVAL: 10000
+      PRIORITY: 200
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      DISABLE_SNAPSHOT: True
+  #
+  EVAL_HOOKS:
+    -
+      NAME: ProbeDataHook
+      PROB_INTERVAL: 100
+      SAVE_LAST: True
+      SAVE_NAME_PREFIX: 'step'
+      SAVE_PROBE_PREFIX: 'image'

scepter/methods/studio/self_train/edit/stylebooth_tb_pro.yaml

@@ -142,13 +142,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: -1
+  RESCALE_LR: False
   #
   WORK_DIR:
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
   #
@@ -258,7 +259,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2
@@ -267,7 +268,7 @@ SOLVER:
   TRAIN_DATA:
     NAME: ImageTextPairMSDataset
     MODE: train
-    MS_DATASET_NAME: cache/save_data/delogo/
+    MS_DATASET_NAME: ./cache/cache_data/delogo/
     MS_DATASET_NAMESPACE: ""
     MS_DATASET_SPLIT: "train"
     MS_DATASET_SUBNAME: ""

scepter/methods/studio/self_train/sd_xl/sdxl_pro.yaml

@@ -147,6 +147,7 @@ SOLVER:
   MAX_EPOCHS: -1
   # NUM_FOLDS DESCRIPTION: Num folds for training. TYPE: int default: 1
   NUM_FOLDS: 1
+  RESCALE_LR: False
   #
   EVAL_INTERVAL: -1
   # WORK_DIR DESCRIPTION: Save dir of the training log or model. TYPE: str default: ''
@@ -155,7 +156,7 @@ SOLVER:
   LOG_FILE: std_log.txt
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   TUNER:
   # MODEL DESCRIPTION:  TYPE:  default: ''
   MODEL:
@@ -538,7 +539,7 @@ SOLVER:
   OPTIMIZER:
     # NAME DESCRIPTION:  TYPE:  default: ''
     NAME: AdamW
-    LEARNING_RATE: 0.0064
+    LEARNING_RATE: 0.00001
     EPS: 1e-8
     AMSGRAD: False
   #

scepter/methods/studio/self_train/self_train.yaml

@@ -13,8 +13,10 @@ TRAIN_PARAS:
     VALUES: [[256, 256], [320, 180], [180, 320],
              [512, 512], [640, 360], [360, 640],
              [768, 768], [960, 540], [540, 960],
-             [1024, 1024], [1280, 720], [720, 1280]]
+             [1024, 1024], [1280, 720], [720, 1280],
+             [720, 480], [480, 720]]
     DEFAULT: [1024, 1024]
   EVAL_PROMPTS:
     - a boy wearing a jacket
     - a dog running on the lawn
+SAVE_FILE_LOCAL_PATH: "cache/scepter_ui/datasets/train_data_from_list"

scepter/methods/studio/self_train/stable_diffusion/sd15_pro.yaml

@@ -137,13 +137,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: -1
+  RESCALE_LR: False
   #
   WORK_DIR:
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
   #
@@ -249,7 +250,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/studio/self_train/stable_diffusion/sd21_pro.yaml

@@ -80,13 +80,14 @@ SOLVER:
   NUM_FOLDS: 1
   ACCU_STEP: 1
   EVAL_INTERVAL: -1
+  RESCALE_LR: False
   #
   WORK_DIR:
   LOG_FILE: std_log.txt
   #
   FILE_SYSTEM:
     NAME: "ModelscopeFs"
-    TEMP_DIR: "./cache/data"
+    TEMP_DIR: "./cache/cache_data"
   #
   FREEZE:
   #
@@ -191,7 +192,7 @@ SOLVER:
   #
   OPTIMIZER:
     NAME: AdamW
-    LEARNING_RATE: 0.064
+    LEARNING_RATE: 0.0001
     BETAS: [ 0.9, 0.999 ]
     EPS: 1e-8
     WEIGHT_DECAY: 1e-2

scepter/methods/studio/tuner_manager/tuner_manager.yaml

@@ -13,3 +13,5 @@ BASE_MODEL_VERSION:
     TUNER_TYPE: [ 'LORA', 'SCE', 'FULL' ]
   - BASE_MODEL: 'EDIT'
     TUNER_TYPE: [ 'LORA', 'SCE', 'FULL' ]
+  - BASE_MODEL: 'FLUX1.0_DEV'
+    TUNER_TYPE: [ 'LORA' ]

scepter/modules/__init__.py

@@ -1,4 +1,23 @@
 # -*- coding: utf-8 -*-
 # Copyright (c) Alibaba, Inc. and its affiliates.
-from scepter.modules import (data, inference, model, opt, solver, transform,
-                             utils)
+from typing import TYPE_CHECKING
+from scepter.modules.utils.import_utils import LazyImportModule
+
+
+if TYPE_CHECKING:
+    from scepter.modules import (data, inference, model, opt, solver, transform,
+                                 utils)
+else:
+    _import_structure = {
+        'modules': ['data', 'inference', 'model', 'opt', 'solver',
+                    'transform', 'utils']
+    }
+
+    import sys
+    sys.modules[__name__] = LazyImportModule(
+        __name__,
+        globals()['__file__'],
+        _import_structure,
+        module_spec=__spec__,
+        extra_objects={},
+    )

scepter/modules/annotator/__init__.py

@@ -1,11 +1,64 @@
 # -*- coding: utf-8 -*-
 # Copyright (c) Alibaba, Inc. and its affiliates.
-from scepter.modules.annotator.base_annotator import GeneralAnnotator
-from scepter.modules.annotator.canny import CannyAnnotator
-from scepter.modules.annotator.color import ColorAnnotator
-from scepter.modules.annotator.hed import HedAnnotator
-from scepter.modules.annotator.identity import IdentityAnnotator
-from scepter.modules.annotator.invert import InvertAnnotator
-from scepter.modules.annotator.midas_op import MidasDetector
-from scepter.modules.annotator.mlsd_op import MLSDdetector
-from scepter.modules.annotator.openpose import OpenposeAnnotator
+from typing import TYPE_CHECKING
+from scepter.modules.utils.import_utils import LazyImportModule
+
+if TYPE_CHECKING:
+    from scepter.modules.annotator.base_annotator import GeneralAnnotator
+    from scepter.modules.annotator.canny import CannyAnnotator
+    from scepter.modules.annotator.color import ColorAnnotator
+    from scepter.modules.annotator.degradation import DegradationAnnotator
+    from scepter.modules.annotator.doodle import DoodleAnnotator
+    from scepter.modules.annotator.gray import GrayAnnotator
+    from scepter.modules.annotator.hed import HedAnnotator
+    from scepter.modules.annotator.identity import IdentityAnnotator
+    from scepter.modules.annotator.informative_drawing import (
+        InfoDrawAnimeAnnotator, InfoDrawContourAnnotator,
+        InfoDrawOpenSketchAnnotator)
+    from scepter.modules.annotator.inpainting import InpaintingAnnotator
+    from scepter.modules.annotator.invert import InvertAnnotator
+    from scepter.modules.annotator.midas_op import MidasDetector
+    from scepter.modules.annotator.mlsd_op import MLSDdetector
+    from scepter.modules.annotator.openpose import OpenposeAnnotator
+    from scepter.modules.annotator.outpainting import OutpaintingAnnotator, OutpaintingResize
+    from scepter.modules.annotator.pidinet import PiDiAnnotator
+    from scepter.modules.annotator.segmentation import ESAMAnnotator
+    from scepter.modules.annotator.sketch import SketchAnnotator
+    from scepter.modules.annotator.lama import LamaAnnotator
+    from scepter.modules.annotator.mask_aug import MaskAugAnnotator, MaskDrawAnnotator, MaskLayoutAnnotator
+    from scepter.modules.annotator.raft import RAFTAnnotator, RAFTVisAnnotator
+else:
+    _import_structure = {
+        'base_annotator': ['GeneralAnnotator'],
+        'canny': ['CannyAnnotator'],
+        'color': ['ColorAnnotator'],
+        'degradation': ['DegradationAnnotator'],
+        'doodle': ['DoodleAnnotator'],
+        'gray': ['GrayAnnotator'],
+        'hed': ['HedAnnotator'],
+        'identity': ['IdentityAnnotator'],
+        'informative_drawing': ['InfoDrawAnimeAnnotator',
+                                'InfoDrawContourAnnotator',
+                                'InfoDrawOpenSketchAnnotator'],
+        'inpainting': ['InpaintingAnnotator'],
+        'invert': ['InvertAnnotator'],
+        'midas_op': ['MidasDetector'],
+        'mlsd_op': ['MLSDdetector'],
+        'openpose': ['OpenposeAnnotator'],
+        'outpainting': ['OutpaintingAnnotator', 'OutpaintingResize'],
+        'pidinet': ['PiDiAnnotator'],
+        'segmentation': ['ESAMAnnotator'],
+        'sketch': ['SketchAnnotator'],
+        'lama': ['LamaAnnotator'],
+        'mask_aug': ['MaskAugAnnotator', 'MaskDrawAnnotator', 'MaskLayoutAnnotator'],
+        'raft': ['RAFTAnnotator', 'RAFTVisAnnotator'],
+    }
+
+    import sys
+    sys.modules[__name__] = LazyImportModule(
+        __name__,
+        globals()['__file__'],
+        _import_structure,
+        module_spec=__spec__,
+        extra_objects={},
+    )