Support loading long clipl model with the CLIP loader node.

comfy/clip_model.py

@@ -88,10 +88,11 @@ class CLIPTextModel_(torch.nn.Module):
         heads = config_dict["num_attention_heads"]
         intermediate_size = config_dict["intermediate_size"]
         intermediate_activation = config_dict["hidden_act"]
+        num_positions = config_dict["max_position_embeddings"]
         self.eos_token_id = config_dict["eos_token_id"]
 
         super().__init__()
-        self.embeddings = CLIPEmbeddings(embed_dim, dtype=dtype, device=device, operations=operations)
+        self.embeddings = CLIPEmbeddings(embed_dim, num_positions=num_positions, dtype=dtype, device=device, operations=operations)
         self.encoder = CLIPEncoder(num_layers, embed_dim, heads, intermediate_size, intermediate_activation, dtype, device, operations)
         self.final_layer_norm = operations.LayerNorm(embed_dim, dtype=dtype, device=device)
 

comfy/sd.py

@@ -24,6 +24,7 @@ import comfy.text_encoders.sa_t5
 import comfy.text_encoders.aura_t5
 import comfy.text_encoders.hydit
 import comfy.text_encoders.flux
+import comfy.text_encoders.long_clipl
 
 import comfy.model_patcher
 import comfy.lora
@@ -443,8 +444,13 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
             clip_target.clip = comfy.text_encoders.sa_t5.SAT5Model
             clip_target.tokenizer = comfy.text_encoders.sa_t5.SAT5Tokenizer
         else:
-            clip_target.clip = sd1_clip.SD1ClipModel
-            clip_target.tokenizer = sd1_clip.SD1Tokenizer
+            w = clip_data[0].get("text_model.embeddings.position_embedding.weight", None)
+            if w is not None and w.shape[0] == 248:
+                clip_target.clip = comfy.text_encoders.long_clipl.LongClipModel
+                clip_target.tokenizer = comfy.text_encoders.long_clipl.LongClipTokenizer
+            else:
+                clip_target.clip = sd1_clip.SD1ClipModel
+                clip_target.tokenizer = sd1_clip.SD1Tokenizer
     elif len(clip_data) == 2:
         if clip_type == CLIPType.SD3:
             clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(clip_l=True, clip_g=True, t5=False)

comfy/text_encoders/long_clipl.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": 49407,
+  "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": 248,
+  "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": "4.24.0",
+  "vocab_size": 49408
+}

comfy/text_encoders/long_clipl.py

@@ -0,0 +1,19 @@
+from comfy import sd1_clip
+import os
+
+class LongClipTokenizer_(sd1_clip.SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(max_length=248, embedding_directory=embedding_directory, tokenizer_data=tokenizer_data)
+
+class LongClipModel_(sd1_clip.SDClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        textmodel_json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "long_clipl.json")
+        super().__init__(device=device, textmodel_json_config=textmodel_json_config, return_projected_pooled=False, dtype=dtype, model_options=model_options)
+
+class LongClipTokenizer(sd1_clip.SD1Tokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, tokenizer=LongClipTokenizer_)
+
+class LongClipModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}, **kwargs):
+        super().__init__(device=device, dtype=dtype, model_options=model_options, clip_model=LongClipModel_, **kwargs)

comfy/text_encoders/sd3_clip.py

@@ -15,7 +15,7 @@ class T5XXLModel(sd1_clip.SDClipModel):
 class T5XXLTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "t5_tokenizer")
-        super().__init__(tokenizer_path, pad_with_end=False, embedding_size=4096, embedding_key='t5xxl', tokenizer_class=T5TokenizerFast, has_start_token=False, pad_to_max_length=False, max_length=99999999, min_length=77)
+        super().__init__(tokenizer_path, embedding_directory=embedding_directory, pad_with_end=False, embedding_size=4096, embedding_key='t5xxl', tokenizer_class=T5TokenizerFast, has_start_token=False, pad_to_max_length=False, max_length=99999999, min_length=77)
 
 
 class SD3Tokenizer: