modified: src/flux/modules/conditioner.py

src/flux/modules/conditioner.py

@@ -14,9 +14,17 @@ class HFEmbedder(nn.Module):
             #self.tokenizer: CLIPTokenizer = CLIPTokenizer.from_pretrained(version, max_length=max_length)
             self.tokenizer: CLIPTokenizer = CLIPTokenizer.from_pretrained("/home/user/app/models/tokenizer", max_length=max_length, truncation=True)
             self.hf_module: CLIPTextModel = CLIPTextModel.from_pretrained(version, **hf_kwargs)
+
+            # --- DEBUG ---
+            print(f"CLIP Tokenizer max length: {CLIPTokenizer.model_max_length}")
+            print(f"CLIP max position embeddings: {CLIPTextModel.config.max_position_embeddings}")
+
         else:
             self.tokenizer: T5Tokenizer = T5Tokenizer.from_pretrained(version, max_length=max_length, truncation=True)
             self.hf_module: T5EncoderModel = T5EncoderModel.from_pretrained(version, **hf_kwargs)
+            # --- DEBUG ---
+            '''print(f"T5 Tokenizer max length: {T5Tokenizer.model_max_length}")
+            print(f"T5 max position embeddings: {T5EncoderModel.config.max_position_embeddings}")'''
 
         self.hf_module = self.hf_module.eval().requires_grad_(False)
 
@@ -42,8 +50,6 @@ class HFEmbedder(nn.Module):
         assert input_ids.shape[1] == self.max_length, f"Sequence length {input_ids.shape[1]} does not match max_length {self.max_length}"
         print(input_ids)
 
-        print(f"self.tokenizer.vocab_size: {self.tokenizer.vocab_size}")  # Debug
-        print(f"self.hf_module.config.vocab_size: {self.hf_module.config.vocab_size}")  # Debug
         print(f"self.tokenizer.vocab_size: {self.tokenizer.vocab_size}")  # Debug
         print(f"self.hf_module.config.vocab_size: {self.hf_module.config.vocab_size}")  # Debug