Update app.py

app.py

@@ -50,34 +50,26 @@ def process_and_describe(payload: dict):
         # 1️⃣ Save & upload image
         local_tmp_path, hf_url, path_in_repo, size_bytes = save_and_upload_image(image_b64)
         
-        # --- NEW VLM CALL LOGIC ---
         # 2️⃣ Prepare prompt in the Qwen specific format (using Markdown for image embedding)
-        # The API expects an image embedded in the prompt using Markdown syntax.
         prompt_text = "Describe this image in detail."
         
-        # Base64 encode the image for embedding in the JSON payload
+        # Base64 encode the image for embedding in the prompt
         with open(local_tmp_path, "rb") as f:
             image_b64_encoded_string = base64.b64encode(f.read()).decode("utf-8")
 
-        # The full prompt format required by Qwen in the API
-        full_prompt = f'<img src="data:image/jpeg;base64,{image_b64_encoded_string}">{prompt_text}'
+        # The full prompt format required by Qwen, embedded in a chat-like structure for the API
+        full_prompt = f'<img src="data:image/jpeg;base64,{image_b64_encoded_string}"> {prompt_text}'
 
-        # 3️⃣ Call VLM using hf_client.post (low-level API call for specific models)
-        # We use the 'text-generation' task endpoint as indicated by the error message.
-        api_response = hf_client.post(
-            json={"inputs": full_prompt, "parameters": {"max_new_tokens": 150}},
+        # 3️⃣ Call VLM using hf_client.text_generation (the preferred method for general LLMs)
+        # This sends the custom prompt string to the model endpoint.
+        vlm_text = hf_client.text_generation(
             model=HF_VLM_MODEL,
-            task="text-generation"
+            prompt=full_prompt,
+            max_new_tokens=150,
+            # Other parameters like temperature can be added here if needed
         )
         
-        # The response is usually a list of dicts, extract the generated text
-        # Example response format: [{'generated_text': '... description ...'}]
-        if isinstance(api_response, list) and len(api_response) > 0:
-            vlm_text = api_response[0].get('generated_text', '').strip()
-        else:
-            vlm_text = "Failed to parse VLM response."
-
-        # --- END NEW VLM CALL LOGIC ---
+        # The response from text_generation is already the cleaned string
 
         return {
             "saved_to_hf_hub": True,
@@ -86,12 +78,12 @@ def process_and_describe(payload: dict):
             "image_url": hf_url,
             "file_size_bytes": size_bytes,
             "robot_id": robot_id,
-            "vlm_description": vlm_text
+            "vlm_description": vlm_text.strip()
         }
 
     except Exception as e:
-        # Added better error handling as suggested previously
-        return {"error": str(e)}
+        # Added better error handling
+        return {"error": f"An API error occurred: {str(e)}"}
 
 # --- Gradio MCP Interface ---
 demo = gr.Interface(