Files

T

Bubbliiiing 403f1f7b78 Fix Bug in Self Forcing in Multi-Gpus Infernece && Update Training Code and Docs && Update Reward Models (#498 )

2026-07-14 10:30:54 +08:00

25 KiB

Executable File

Raw Permalink Blame History

Qwen-Image-Edit Full Parameter Training Guide

This document provides a complete workflow for full parameter training of Qwen-Image-Edit Diffusion Transformer, including environment configuration, data preparation, distributed training, and inference testing.

1. Environment Configuration
2. Data Preparation
3. Full Parameter Training
4. Inference Testing
5. Additional Resources

1. Environment Configuration

Option 1: Using requirements.txt

pip install -r requirements.txt

Option 2: Manual Installation

pip install Pillow einops safetensors timm tomesd librosa "torch>=2.1.2" torchdiffeq torchsde decord datasets numpy scikit-image
pip install omegaconf SentencePiece imageio[ffmpeg] imageio[pyav] tensorboard beautifulsoup4 ftfy func_timeout onnxruntime
pip install "peft>=0.17.0" "accelerate>=0.25.0" "gradio>=3.41.2" "diffusers>=0.30.1" "transformers>=4.46.2"
pip install yunchang xfuser modelscope openpyxl
pip uninstall opencv-python opencv-contrib-python opencv-python-headless -y
pip install opencv-python-headless
pip install deepspeed==0.17.0 numpy==1.26.4

Option 3: Using Docker

When using Docker, please ensure that the graphics driver and CUDA environment are correctly installed, then execute the following commands:

# pull image
docker pull mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun

# enter image
docker run -it -p 7860:7860 --network host --gpus all --security-opt seccomp:unconfined --shm-size 200g mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun

2. Data Preparation

2.1 Quick Test Dataset

We provide a test dataset containing several training samples.

# Download official example dataset
modelscope download --dataset PAI/X-Fun-Images-Edit-Demo --local_dir ./datasets/X-Fun-Images-Edit-Demo

2.2 Dataset Structure

📦 datasets/
├── 📂 my_dataset/
│   ├── 📂 train/
│   │   ├── 📄 image001.jpg
│   │   ├── 📄 image002.png
│   │   └── 📄 ...
│   └── 📄 metadata.json

2.3 metadata.json Format

Edit Model Data Format:

The metadata.json for Edit model is different from the normal version, requiring the addition of a source_file_path field.

Qwen-Image-Edit: Only needs one file in source_file_path
Qwen-Image-Edit-2509: Needs one or more files in source_file_path

Relative Path Format (Edit Model):

[
    {
      "file_path": "train/00000001.jpg",
      "source_file_path": ["source/00000001.jpg"],
      "text": "A young woman stands on a sunny coastline, wearing a refreshing white shirt and skirt",
      "type": "image"
    },
    {
      "file_path": "train/00000002.jpg",
      "source_file_path": ["source/00000002.jpg"],
      "text": "A young woman with purple hair stands on the coastline, with the vast sea in the background",
      "type": "image"
    }
]

Key Fields Description:

file_path: Target image path (the image to be generated after training)
source_file_path: Source image path array (original images for editing)
- Edit model will edit based on source_file_path images according to text description to generate file_path images
- Qwen-Image-Edit only needs one source file, Qwen-Image-Edit-2509 supports multiple source files
text: Image description (prompt, describing the expected generated content)
type: Data type ("image" or "video")
width / height: Image width and height (recommended to provide, used for bucket training. If not provided, it will be automatically read during training, which may affect training speed when data is stored on slow systems like OSS).
- You can use scripts/process_json_add_width_and_height.py to extract width and height fields from json without these fields, supporting both images and videos.
- Usage: python scripts/process_json_add_width_and_height.py --input_file datasets/X-Fun-Images-Demo/metadata.json --output_file datasets/X-Fun-Images-Demo/metadata_add_width_height.json.

2.4 Relative vs Absolute Path Usage

Relative Paths:

If your data uses relative paths, set in the training script:

export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"

Absolute Paths:

If your data uses absolute paths, set in the training script:

export DATASET_NAME=""
export DATASET_META_NAME="/mnt/data/metadata_add_width_height.json"

💡 Recommendation: If the dataset is small and stored locally, relative paths are recommended; if the dataset is stored on external storage (such as NAS, OSS) or shared across multiple machines, absolute paths are recommended.

3. Full Parameter Training

3.1 Download Pre-trained Model

# Create model directory
mkdir -p models/Diffusion_Transformer

# Download official Qwen-Image-Edit weights
modelscope download --model Qwen/Qwen-Image-Edit --local_dir models/Diffusion_Transformer/Qwen-Image-Edit

3.2 Quick Start (DeepSpeed-Zero-2)

If you have downloaded data according to 2.1 Quick Test Dataset and downloaded weights according to 3.1 Download Pre-trained Model, you can directly copy the quick start command to launch.

Training Notes:

Warning Without DeepSpeed: Training Qwen-Image-Edit without DeepSpeed may result in insufficient GPU memory. DeepSpeed-Zero-2 or FSDP is recommended.
DeepSpeed Zero-3 Recommendation: DeepSpeed Zero-3 is not highly recommended at the moment. In this repository, using FSDP has fewer errors and is more stable.

If using DeepSpeed Zero-3, after training you need to use the following command to get the final model:

python scripts/zero_to_bf16.py output_dir/checkpoint-{step-number} output_dir/checkpoint-{step-number}-outputs --max_shard_size 80GB --safe_serialization

DeepSpeed-Zero-2 and FSDP are recommended for training. Here we use DeepSpeed-Zero-2 as an example to configure the shell file.

The difference between DeepSpeed-Zero-2 and FSDP in this document is whether the model weights are sharded. If using multiple GPUs and encountering insufficient memory with DeepSpeed-Zero-2, you can switch to FSDP for training.

export MODEL_NAME="models/Diffusion_Transformer/Qwen-Image-Edit"
export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --use_deepspeed --deepspeed_config_file config/zero_stage2_config.json --deepspeed_multinode_launcher standard scripts/qwenimage/train_edit.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_qwenimage_edit" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --uniform_sampling \
  --trainable_modules "." \
  --train_mode "qwen_image_edit"

3.3 Training Parameters Explanation

Key Parameters Description:

Parameter	Description	Example Value
`--pretrained_model_name_or_path`	Pre-trained model path	`models/Diffusion_Transformer/Qwen-Image-Edit`
`--train_data_dir`	Training data directory	`datasets/X-Fun-Images-Edit-Demo/`
`--train_data_meta`	Training data metadata file	`datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json`
`--train_batch_size`	Number of samples per batch	1
`--image_sample_size`	Maximum training resolution, code will automatically bucket	1328
`--gradient_accumulation_steps`	Gradient accumulation steps (equivalent to increasing batch)	1
`--dataloader_num_workers`	Number of DataLoader subprocesses	8
`--num_train_epochs`	Number of training epochs	100
`--checkpointing_steps`	Save checkpoint every N steps	50
`--learning_rate`	Initial learning rate	2e-05
`--lr_scheduler`	Learning rate scheduler	`constant_with_warmup`
`--lr_warmup_steps`	Learning rate warmup steps	100
`--seed`	Random seed	42
`--output_dir`	Output directory	`output_dir`
`--gradient_checkpointing`	Activation recomputation	-
`--mixed_precision`	Mixed precision: `fp16/bf16`	`bf16`
`--adam_weight_decay`	AdamW weight decay	3e-2
`--adam_epsilon`	AdamW epsilon value	1e-10
`--vae_mini_batch`	Mini batch size for VAE encoding	1
`--max_grad_norm`	Gradient clipping threshold	0.05
`--enable_bucket`	Enable bucket training, no image cropping, train entire images grouped by resolution	-
`--random_hw_adapt`	Auto-scale images to random sizes in range `[512, image_sample_size]`	-
`--resume_from_checkpoint`	Resume training path, use `"latest"` to auto-select latest checkpoint	None
`--uniform_sampling`	Uniform sampling of timesteps	-
`--trainable_modules`	Trainable modules (`.` means all modules)	`"."`
`--train_mode`	Training mode: `qwen_image_edit` for Qwen-Image-Edit, `qwen_image_edit_plus` for Qwen-Image-Edit-2509	`"qwen_image_edit"`
`--validation_steps`	Run validation every N steps	100
`--validation_epochs`	Run validation every N epochs	500
`--validation_prompts`	Prompts used during validation	`"1girl, black_hair, ..."`
`--validation_image_paths`	Source image paths used during validation (Edit model specific)	`"asset/8.png"`

random_hw_adapt Detailed Explanation:

When random_hw_adapt is enabled and image_sample_size=1024, the resolution range of training images is 512x512 to 1024x1024
Can be used with enable_bucket for more flexible handling of different image resolutions
For example: random_hw_adapt=true, image_sample_size=1024, images will be randomly scaled to sizes between 512 and 1024 during training

3.4 Training Validation

You can configure validation parameters to regularly generate test images during training to monitor training progress and model quality.

Validation Parameters Configuration:

accelerate launch --use_deepspeed --deepspeed_config_file config/zero_stage2_config.json --deepspeed_multinode_launcher standard scripts/qwenimage/train_edit.py \
  # ... (other training parameters)
  --validation_steps=100 \
  --validation_epochs=500 \
  --validation_prompts="1girl, black_hair, brown_eyes, earrings, freckles, grey_background, jewelry, lips, long_hair, looking_at_viewer, nose, piercing, realistic, red_lips, solo, upper_body" \
  --validation_image_paths="asset/8.png"

Parameters Description:

Parameter	Description	Recommended Value
`--validation_steps`	Run validation every N steps. If dataset is large and you want to save validation time, you can set a larger value (e.g., 100 or 500)	100
`--validation_epochs`	Run validation every N epochs	500
`--validation_prompts`	Prompts for validation image generation. Multiple prompts can be set, separated by spaces	Multiple space-separated prompts
`--validation_image_paths`	Source image paths used during validation (Edit model specific)	`asset/8.png`

Notes:

Validation images will be saved to the output_dir directory
Setting --validation_steps=1 means validation at every step, which may slow down training. Adjust according to actual needs.
Multi-prompt validation format: --validation_prompts "prompt1" "prompt2" "prompt3"
Edit model validation must provide --validation_image_paths parameter to specify source images for editing

3.5 Training with FSDP

If using multiple GPUs and encountering insufficient memory with DeepSpeed-Zero-2, you can switch to FSDP for training.

export MODEL_NAME="models/Diffusion_Transformer/Qwen-Image-Edit"
export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --mixed_precision="bf16" --use_fsdp --fsdp_auto_wrap_policy TRANSFORMER_BASED_WRAP --fsdp_transformer_layer_cls_to_wrap=QwenImageTransformerBlock --fsdp_sharding_strategy "FULL_SHARD" --fsdp_state_dict_type=SHARDED_STATE_DICT --fsdp_backward_prefetch "BACKWARD_PRE" --fsdp_cpu_ram_efficient_loading False scripts/qwenimage/train_edit.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_qwenimage_edit" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --uniform_sampling \
  --trainable_modules "." \
  --train_mode "qwen_image_edit"

3.6 Other Backends

3.6.1 Training with DeepSpeed-Zero-3

DeepSpeed Zero-3 is not highly recommended at the moment. In this repository, using FSDP has fewer errors and is more stable.

DeepSpeed Zero-3:

After training, you can use the following command to get the final model:

python scripts/zero_to_bf16.py output_dir/checkpoint-{step-number} output_dir/checkpoint-{step-number}-outputs --max_shard_size 80GB --safe_serialization

Training shell command is as follows:

export MODEL_NAME="models/Diffusion_Transformer/Qwen-Image-Edit"
export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --zero_stage 3 --zero3_save_16bit_model true --zero3_init_flag true --use_deepspeed --deepspeed_config_file config/zero_stage3_config.json --deepspeed_multinode_launcher standard scripts/qwenimage/train_edit.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --uniform_sampling \
  --trainable_modules "." \
  --train_mode "qwen_image_edit"

3.6.2 Training Without DeepSpeed and FSDP

This approach is not recommended because without memory-saving backends, it easily causes insufficient GPU memory. This is only provided as a reference shell for training.

export MODEL_NAME="models/Diffusion_Transformer/Qwen-Image-Edit"
export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --mixed_precision="bf16" scripts/qwenimage/train_edit.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_qwenimage_edit" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --uniform_sampling \
  --trainable_modules "." \
  --train_mode "qwen_image_edit"

3.7 Multi-machine Distributed Training

Suitable for: Ultra-large-scale datasets, faster training speed

3.7.1 Environment Configuration

Assuming 2 machines, each with 8 GPUs:

Machine 0 (Master):

export MODEL_NAME="models/Diffusion_Transformer/Qwen-Image-Edit"
export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"
export MASTER_ADDR="192.168.1.100"  # Master machine IP
export MASTER_PORT=10086
export WORLD_SIZE=2                  # Total number of machines
export NUM_PROCESS=16                # Total processes = machines × 8
export RANK=0                        # Current machine rank (0 or 1)
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --mixed_precision="bf16" --main_process_ip=$MASTER_ADDR --main_process_port=$MASTER_PORT --num_machines=$WORLD_SIZE --num_processes=$NUM_PROCESS --machine_rank=$RANK --use_deepspeed --deepspeed_config_file config/zero_stage2_config.json --deepspeed_multinode_launcher standard scripts/qwenimage/train_edit.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_qwenimage_edit" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --uniform_sampling \
  --trainable_modules "." \
  --train_mode "qwen_image_edit"

Machine 1 (Worker):

export MODEL_NAME="models/Diffusion_Transformer/Qwen-Image-Edit"
export DATASET_NAME="datasets/X-Fun-Images-Edit-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-Edit-Demo/metadata_add_width_height.json"
export MASTER_ADDR="192.168.1.100"  # Same as Master
export MASTER_PORT=10086
export WORLD_SIZE=2
export NUM_PROCESS=16
export RANK=1  # Note this is 1
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

# Use the same accelerate launch command as Machine 0

3.7.2 Multi-machine Training Notes

Network Requirements:
- RDMA/InfiniBand recommended (high performance)
- Without RDMA, add environment variables:
```
export NCCL_IB_DISABLE=1
export NCCL_P2P_DISABLE=1
```
Data Synchronization: All machines must be able to access the same data paths (NFS/shared storage)

4. Inference Testing

4.1 Inference Parameters Explanation

Key Parameters Description:

Parameter	Description	Example Value
`GPU_memory_mode`	Memory management mode, see table below for options	`model_group_offload`
`ulysses_degree`	Head dimension parallelism, 1 for single GPU	1
`ring_degree`	Sequence dimension parallelism, 1 for single GPU	1
`fsdp_dit`	Use FSDP for Transformer during multi-GPU inference to save memory	`False`
`fsdp_text_encoder`	Use FSDP for text encoder during multi-GPU inference	`False`
`compile_dit`	Compile Transformer for faster inference (effective at fixed resolution)	`False`
`enable_teacache`	Enable TeaCache for faster inference	`True`
`teacache_threshold`	TeaCache threshold, recommended 0.05~0.30, larger is faster but quality may decrease	0.25
`num_skip_start_steps`	Steps to skip at the beginning of inference to reduce impact on generation quality	5
`teacache_offload`	Offload TeaCache tensors to CPU to save memory	`False`
`cfg_skip_ratio`	Skip some CFG steps for faster inference, recommended 0.00~0.25	0
`model_name`	Model path	`models/Diffusion_Transformer/Qwen-Image-Edit`
`sampler_name`	Sampler type: `Flow`, `Flow_Unipc`, `Flow_DPM++`	`Flow`
`transformer_path`	Path to load trained Transformer weights	`None`
`vae_path`	Path to load trained VAE weights	`None`
`lora_path`	LoRA weights path	`None`
`sample_size`	Generated image resolution `[height, width]`	`[1344, 768]`
`weight_dtype`	Model weight precision, use `torch.float16` for GPUs that don't support bf16	`torch.bfloat16`
`prompt`	Positive prompt, describes content to generate	`"1girl, black_hair..."`
`negative_prompt`	Negative prompt, content to avoid	`" "`
`guidance_scale`	Guidance strength	4.0
`seed`	Random seed, for reproducibility	43
`num_inference_steps`	Inference steps	50
`lora_weight`	LoRA weight intensity	0.55
`save_path`	Generated image save path	`samples/qwenimage-t2i`

Memory Management Mode Description:

Mode	Description	Memory Usage
`model_full_load`	Entire model loaded to GPU	Highest
`model_full_load_and_qfloat8`	Full load + FP8 quantization	High
`model_cpu_offload`	Offload model to CPU after use	Medium
`model_cpu_offload_and_qfloat8`	CPU offload + FP8 quantization	Medium-Low
`model_group_offload`	Layer groups switch between CPU/CUDA	Low
`sequential_cpu_offload`	Layer-by-layer offload (slowest)	Lowest

4.2 Single GPU Inference

Quick Start

Run single GPU inference with the following command:

python examples/qwenimage/predict_t2i_edit.py

Edit examples/qwenimage/predict_t2i_edit.py according to your needs. For first-time inference, focus on the following parameters. If you're interested in other parameters, please refer to the inference parameters explanation above.

# Select based on GPU memory
GPU_memory_mode = "model_group_offload"
# Based on actual model path
model_name = "models/Diffusion_Transformer/Qwen-Image-Edit"  
# Trained weights path, e.g., "output_dir_qwenimage_edit/checkpoint-xxx/diffusion_pytorch_model.safetensors"
transformer_path = None  
# Write based on content to generate
prompt = "A young woman stands on a sunny coastline, wearing a refreshing white shirt and skirt"  
# ...

4.3 Multi-GPU Parallel Inference

Suitable for: High-resolution generation, accelerated inference

Install Parallel Inference Dependencies

pip install xfuser==0.4.2 yunchang==0.6.2

Configure Parallel Strategy

Edit examples/qwenimage/predict_t2i_edit.py:

# Ensure ulysses_degree × ring_degree = number of GPUs
# For example, using 2 GPUs:
ulysses_degree = 2  # Head dimension parallelism
ring_degree = 1     # Sequence dimension parallelism

Configuration Principles:

ulysses_degree must be divisible by the model's head count.
ring_degree splits on sequence dimension, affecting communication overhead. Try not to use it when head count can be split.

Example Configurations:

GPU Count	ulysses_degree	ring_degree	Description
1	1	1	Single GPU
4	4	1	Head parallelism
8	8	1	Head parallelism
8	4	2	Hybrid parallelism

Run Multi-GPU Inference

torchrun --nproc-per-node=2 examples/qwenimage/predict_t2i_edit.py

5. Additional Resources

Official GitHub: https://github.com/aigc-apps/VideoX-Fun

25 KiB Executable File Raw Permalink Blame History Unescape Escape