Upload LLM2Vec4CXR fine-tuned model

README.md

@@ -82,6 +82,9 @@ tokenizer.padding_side = 'left'
 # Example usage for chest X-ray report analysis
 def encode_text(text):
     inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
+    # IMPORTANT: Add embed_mask for proper model functioning
+    # For simple text encoding, embed_mask is the same as attention_mask
+    inputs["embed_mask"] = inputs["attention_mask"].clone()
     with torch.no_grad():
         embeddings = model(inputs)
     return embeddings
@@ -91,6 +94,8 @@ report = "There is a small increase in the left-sided effusion. There continues
 embedding = encode_text(report)
 ```
 
+**Note**: The model requires an `embed_mask` input. For simple text encoding, set `embed_mask` equal to `attention_mask`. For instruction-following tasks, use the separator-based tokenization shown below.
+
 ### Advanced Usage with Separator-based Processing
 
 The model supports special separator-based processing for instruction-following tasks:

usage_example.py

@@ -162,6 +162,8 @@ def main():
     report = "There is a small increase in the left-sided effusion. There continues to be volume loss at both bases."
     
     inputs = tokenizer(report, return_tensors="pt", padding=True, truncation=True, max_length=512)
+    # Add embed_mask (same as attention_mask for simple text encoding)
+    inputs["embed_mask"] = inputs["attention_mask"].clone()
     inputs = inputs.to(device)
     
     with torch.no_grad():
@@ -223,6 +225,8 @@ def main():
     
     print("Computing embeddings for multiple reports...")
     inputs = tokenizer(reports, return_tensors="pt", padding=True, truncation=True, max_length=512)
+    # Add embed_mask (same as attention_mask for simple text encoding)
+    inputs["embed_mask"] = inputs["attention_mask"].clone()
     inputs = inputs.to(device)
     
     with torch.no_grad():