Revert "Revert "Load clip models from the SDXL checkpoint""

This reverts commit 0891b2db7f.

configs/clip_vit_config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "openai/clip-vit-large-patch14",
+  "architectures": [
+    "CLIPTextModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "eos_token_id": 2,
+  "hidden_act": "quick_gelu",
+  "hidden_size": 768,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "projection_dim": 768,
+  "torch_dtype": "float32",
+  "transformers_version": "4.22.0.dev0",
+  "vocab_size": 49408
+}

configs/tokenizer/config.json

@@ -0,0 +1,171 @@
+{
+  "_name_or_path": "clip-vit-large-patch14/",
+  "architectures": [
+    "CLIPModel"
+  ],
+  "initializer_factor": 1.0,
+  "logit_scale_init_value": 2.6592,
+  "model_type": "clip",
+  "projection_dim": 768,
+  "text_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "bos_token_id": 0,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 2,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "quick_gelu",
+    "hidden_size": 768,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 77,
+    "min_length": 0,
+    "model_type": "clip_text_model",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 12,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 12,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 1,
+    "prefix": null,
+    "problem_type": null,
+    "projection_dim" : 768,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.16.0.dev0",
+    "use_bfloat16": false,
+    "vocab_size": 49408
+  },
+  "text_config_dict": {
+    "hidden_size": 768,
+    "intermediate_size": 3072,
+    "num_attention_heads": 12,
+    "num_hidden_layers": 12,
+    "projection_dim": 768
+  },
+  "torch_dtype": "float32",
+  "transformers_version": null,
+  "vision_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "quick_gelu",
+    "hidden_size": 1024,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "image_size": 224,
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "clip_vision_model",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 16,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 24,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": null,
+    "patch_size": 14,
+    "prefix": null,
+    "problem_type": null,
+    "projection_dim" : 768,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.16.0.dev0",
+    "use_bfloat16": false
+  },
+  "vision_config_dict": {
+    "hidden_size": 1024,
+    "intermediate_size": 4096,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 24,
+    "patch_size": 14,
+    "projection_dim": 768
+  }
+}

configs/tokenizer/preprocessor_config.json

@@ -0,0 +1,19 @@
+{
+  "crop_size": 224,
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "resample": 3,
+  "size": 224
+}

configs/tokenizer/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": {"content": "<|startoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "eos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "unk_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "pad_token": "<|endoftext|>"}

configs/tokenizer/tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+    "unk_token": {
+        "content": "<|endoftext|>",
+        "single_word": false,
+        "lstrip": false,
+        "rstrip": false,
+        "normalized": true,
+        "__type": "AddedToken"
+    },
+    "bos_token": {
+        "content": "<|startoftext|>",
+        "single_word": false,
+        "lstrip": false,
+        "rstrip": false,
+        "normalized": true,
+        "__type": "AddedToken"
+    },
+    "eos_token": {
+        "content": "<|endoftext|>",
+        "single_word": false,
+        "lstrip": false,
+        "rstrip": false,
+        "normalized": true,
+        "__type": "AddedToken"
+    },
+    "pad_token": "<|endoftext|>",
+    "add_prefix_space": false,
+    "errors": "replace",
+    "do_lower_case": true,
+    "name_or_path": "openai/clip-vit-base-patch32",
+    "model_max_length": 77,
+    "special_tokens_map_file": "./special_tokens_map.json",
+    "tokenizer_class": "CLIPTokenizer"
+}

nodes.py

@@ -12,6 +12,13 @@ import torch.cuda
 from .sgm.util import instantiate_from_config
 from .SUPIR.util import convert_dtype, load_state_dict
 
+from open_clip import CLIP, CLIPTextCfg
+from transformers import (
+    CLIPTextModel,
+    CLIPTokenizer,
+    CLIPTextConfig,
+
+)
 script_directory = os.path.dirname(os.path.abspath(__file__))
 
 try:
@@ -22,6 +29,41 @@ try:
 except:
     XFORMERS_IS_AVAILABLE = False
 
+
+def build_text_model_from_openai_state_dict(
+        state_dict: dict,
+        cast_dtype=torch.float16,
+):
+
+    embed_dim = state_dict["text_projection"].shape[1]
+    context_length = state_dict["positional_embedding"].shape[0]
+    vocab_size = state_dict["token_embedding.weight"].shape[0]
+    transformer_width = state_dict["ln_final.weight"].shape[0]
+    transformer_heads = transformer_width // 64
+    transformer_layers = len(set(k.split(".")[2] for k in state_dict if k.startswith(f"transformer.resblocks")))
+
+    vision_cfg = None
+    text_cfg = CLIPTextCfg(
+        context_length=context_length,
+        vocab_size=vocab_size,
+        width=transformer_width,
+        heads=transformer_heads,
+        layers=transformer_layers,
+    )
+    model = CLIP(
+        embed_dim,
+        vision_cfg=vision_cfg,
+        text_cfg=text_cfg,
+        quick_gelu=True,  # OpenAI models were trained with QuickGELU
+        cast_dtype=cast_dtype,
+    )
+
+    model.load_state_dict(state_dict, strict=False)
+    model = model.eval()
+    for param in model.parameters():
+        param.requires_grad = False
+    return model
+
 class SUPIR_Upscale:
     upscale_methods = ["nearest-exact", "bilinear", "area", "bicubic", "lanczos"]
 
@@ -94,7 +136,6 @@ class SUPIR_Upscale:
                 control_scale, cfg_scale_start, control_scale_start, restoration_scale, keep_model_loaded,
                 a_prompt, n_prompt, sdxl_model, supir_model, use_tiled_vae, use_tiled_sampling=False, sampler_tile_size=128, sampler_tile_stride=64, captions="", diffusion_dtype="auto",
                 encoder_dtype="auto", batch_size=1):
-
         device = mm.get_torch_device()
         mm.unload_all_models()
 
@@ -103,6 +144,8 @@ class SUPIR_Upscale:
 
         config_path = os.path.join(script_directory, "options/SUPIR_v0.yaml")
         config_path_tiled = os.path.join(script_directory, "options/SUPIR_v0_tiled.yaml")
+        clip_config_path = os.path.join(script_directory, "configs/clip_vit_config.json")
+        tokenizer_path = os.path.join(script_directory, "configs/tokenizer")
 
         custom_config = {
             'sdxl_model': sdxl_model,
@@ -172,9 +215,11 @@ class SUPIR_Upscale:
             config.model.params.diffusion_dtype = model_dtype
 
             self.model = instantiate_from_config(config.model).cpu()
+
             try:
                 print(f'Attempting to load SUPIR model: [{SUPIR_MODEL_PATH}]')
                 supir_state_dict = load_state_dict(SUPIR_MODEL_PATH)
+                
             except:
                 raise Exception("Failed to load SUPIR model")
             try:
@@ -185,11 +230,43 @@ class SUPIR_Upscale:
             self.model.load_state_dict(supir_state_dict, strict=False)
             self.model.load_state_dict(sdxl_state_dict, strict=False)
 
-            del supir_state_dict, sdxl_state_dict
+            del supir_state_dict
+
+            #first clip model from SDXL checkpoint
+            try:
+                print("Loading first clip model from SDXL checkpoint")
+                
+                replace_prefix = {}
+                replace_prefix["conditioner.embedders.0.transformer."] = ""
+    
+                sd = comfy.utils.state_dict_prefix_replace(sdxl_state_dict, replace_prefix, filter_keys=False)
+                clip_text_config = CLIPTextConfig.from_pretrained(clip_config_path)
+                self.model.conditioner.embedders[0].tokenizer = CLIPTokenizer.from_pretrained(tokenizer_path)
+                self.model.conditioner.embedders[0].transformer = CLIPTextModel(clip_text_config)
+                self.model.conditioner.embedders[0].transformer.load_state_dict(sd, strict=False)
+                self.model.conditioner.embedders[0].eval()
+                for param in self.model.conditioner.embedders[0].parameters():
+                    param.requires_grad = False
+            except:
+                raise Exception("Failed to load first clip model from SDXL checkpoint")
+            
+            del sdxl_state_dict
+
+            #second clip model from SDXL checkpoint
+            try:
+                print("Loading second clip model from SDXL checkpoint")
+                replace_prefix2 = {}
+                replace_prefix2["conditioner.embedders.1.model."] = ""
+                sd = comfy.utils.state_dict_prefix_replace(sd, replace_prefix2, filter_keys=True)                
+                clip_g = build_text_model_from_openai_state_dict(sd, cast_dtype=dtype)
+                self.model.conditioner.embedders[1].model = clip_g
+            except:
+                raise Exception("Failed to load second clip model from SDXL checkpoint")
+        
+            del sd, clip_g
             mm.soft_empty_cache()
 
             try:
-                # to dtype first then to device to reduce memory usage
                 self.model.to(dtype)
                 self.model.to(device)
             except Exception as e:
@@ -209,9 +286,9 @@ class SUPIR_Upscale:
         B, H, W, C = image.shape
         new_height = H // 64 * 64
         new_width = W // 64 * 64
-        image = image.permute(0, 3, 1, 2).contiguous().to(device)
+        image = image.permute(0, 3, 1, 2).contiguous()
         resized_image = F.interpolate(image, size=(new_height, new_width), mode='bicubic', align_corners=False)
-
+        resized_image = resized_image.to(device)
         captions_list = []
         captions_list.append(captions)
         print("captions: ", captions_list)
@@ -260,11 +337,11 @@ class SUPIR_Upscale:
             out_stacked = torch.cat(out, dim=0).cpu().to(torch.float32).permute(0, 2, 3, 1)
         else:
             out_stacked = torch.stack(out, dim=0).cpu().to(torch.float32).permute(0, 2, 3, 1)
+            
         final_image, = ImageScale.upscale(self, out_stacked, "lanczos", W, H, crop="disabled")
 
         return (final_image,)
 
-
 NODE_CLASS_MAPPINGS = {
     "SUPIR_Upscale": SUPIR_Upscale
 }

sgm/modules/encoders/modules.py

@@ -461,12 +461,17 @@ class FrozenCLIPEmbedder(AbstractEmbModel):
     ):  # clip-vit-base-patch32
         super().__init__()
         assert layer in self.LAYERS
-        self.tokenizer = CLIPTokenizer.from_pretrained(version if SDXL_CLIP1_PATH is None else SDXL_CLIP1_PATH)
-        self.transformer = CLIPTextModel.from_pretrained(version if SDXL_CLIP1_PATH is None else SDXL_CLIP1_PATH)
+        #self.tokenizer = CLIPTokenizer.from_pretrained(version if SDXL_CLIP1_PATH is None else SDXL_CLIP1_PATH)
+        #self.transformer = CLIPTextModel.from_pretrained(version if SDXL_CLIP1_PATH is None else SDXL_CLIP1_PATH)
+        #self.clip_text_config = CLIPTextConfig.from_pretrained("openai/clip-vit-large-patch14")
+        #self.tokenizer = CLIPTokenizer.from_pretrained(version)
+        self.tokenizer = None
+        #self.transformer = CLIPTextModel(self.clip_text_config)
+        self.transformer = None
         self.device = device
         self.max_length = max_length
-        if freeze:
-            self.freeze()
+        #if freeze:
+        #    self.freeze()
         self.layer = layer
         self.layer_idx = layer_idx
         self.return_pooled = always_return_pooled
@@ -529,19 +534,20 @@ class FrozenOpenCLIPEmbedder2(AbstractEmbModel):
     ):
         super().__init__()
         assert layer in self.LAYERS
-        model, _, _ = open_clip.create_model_and_transforms(
-            arch,
-            device=torch.device("cpu"),
-            pretrained=version if SDXL_CLIP2_CKPT_PTH is None else SDXL_CLIP2_CKPT_PTH,
-        )
-        del model.visual
-        self.model = model
+        # model, _, _ = open_clip.create_model_and_transforms(
+        #      arch,
+        #      device=torch.device("cpu"),
+        #      pretrained=version if SDXL_CLIP2_CKPT_PTH is None else SDXL_CLIP2_CKPT_PTH,
+        #  )
         
+        # del model.visual
+        # self.model = model
+        self.model = None
         self.device = device
         self.max_length = max_length
         self.return_pooled = always_return_pooled
-        if freeze:
-            self.freeze()
+        #if freeze:
+        #    self.freeze()
         self.layer = layer
         if self.layer == "last":
             self.layer_idx = 0