Update app.py

app.py

@@ -14,32 +14,44 @@ def gradio_predict(input_text):
     tokenized_input = tokenizer(
         input_text,
         return_tensors="np",
-        padding="max_length",
-        truncation=True,  # Ensures input is truncated if it's too long
-        max_length=512  # Ensure the sequence doesn't exceed the model's max length
+        padding='max_length',  # Pad to max length
+        truncation=True,       # Truncate if longer than max length
+        max_length=512         # Ensure the sequence doesn't exceed the model's max length
     )
 
     # Convert tokenized inputs to numpy arrays and ensure correct shape
-    input_ids = np.array(tokenized_input["input_ids"], dtype=np.int64)  # No need for reshape
-    attention_mask = np.array(tokenized_input["attention_mask"], dtype=np.int64)
-    decoder_input_ids = input_ids  # Same as input_ids for translation models
+    input_ids = np.array(tokenized_input["input_ids"], dtype=np.int64)  # Shape should be [1, 512]
+    attention_mask = np.array(tokenized_input["attention_mask"], dtype=np.int64)  # Shape should be [1, 512]
 
-    # Check if arrays are correctly formed
+    # Prepare decoder input ids if required by your model
+    decoder_input_ids = input_ids  # Adjust as needed based on model requirements
+
+    # Debugging output
+    print(f"Input Text: {input_text}")
+    print(f"Tokens: {tokenizer.tokenize(input_text)}")
     print(f"input_ids shape: {input_ids.shape}, attention_mask shape: {attention_mask.shape}")
     
     # Perform inference with ONNX model
-    outputs = session.run(
-        None,
-        {
-            "input_ids": input_ids,
-            "attention_mask": attention_mask,
-            "decoder_input_ids": decoder_input_ids
-        }
-    )
+    try:
+        outputs = session.run(
+            None,
+            {
+                "input_ids": input_ids,
+                "attention_mask": attention_mask,
+                "decoder_input_ids": decoder_input_ids  # Include this only if required by your model
+            }
+        )
+        
+        # Debugging output for outputs
+        print(f"Outputs: {outputs}, type: {type(outputs)}")
+
+        # Decode output and return translated text
+        translated_text = tokenizer.decode(outputs[0][0], skip_special_tokens=True)
+        return translated_text
 
-    # Decode output and return translated text
-    translated_text = tokenizer.decode(outputs[0][0], skip_special_tokens=True)
-    return translated_text
+    except Exception as e:
+        print(f"Error during inference: {e}")
+        return "An error occurred during inference."
 
 # Gradio interface for the web app
 gr.Interface(