Update app.py

app.py

@@ -1,57 +1,36 @@
 import gradio as gr
 import json
 import base64
-from io import BytesIO
 import requests
 import os
 
 # HF token & model
 HF_TOKEN = os.environ.get("HF_CV_ROBOT_TOKEN")
-MODEL = "Qwen/Qwen2.5-VL-7B-Instruct"  # HF 支援列表裡的模型
+MODEL = "Qwen/Qwen2.5-VL-7B-Instruct"
 
 if not HF_TOKEN:
     print("ERROR: HF_CV_ROBOT_TOKEN environment variable not set.")
 
 # -------------------------------
-# 主處理函數 (Main Processing Function)
+# Main Processing Function
 # -------------------------------
 def process(payload: dict):
     try:
         if not HF_TOKEN:
-            return {"error": "Hugging Face token is missing. Please check Space secrets."}
+            return {"error": "Hugging Face token missing. Check environment variable."}
 
         robot_id = payload.get("robot_id", "unknown")
         image_b64 = payload["image_b64"]
 
-        # Base64 -> bytes -> 保存為 tmp.jpg
-        tmp_path = "tmp.jpg"
-        with open(tmp_path, "wb") as f:
-            f.write(base64.b64decode(image_b64))
-
-        # 上傳 image file 到 HF Router
-        files = {"file": open(tmp_path, "rb")}
-        upload_resp = requests.post(
-            "https://huggingface.co/api/uploads",
-            headers={"Authorization": f"Bearer {HF_TOKEN}"},
-            files=files
-        )
-        files["file"].close()
-        os.remove(tmp_path)
-
-        if upload_resp.status_code != 200:
-            return {"error": f"HF upload failed: {upload_resp.status_code}, {upload_resp.text}"}
-
-        file_info = upload_resp.json()
-        file_url = file_info.get("href")  # 取得 HF hosted file URL
-
-        # JSON payload 放文字訊息 + image file reference
+        # 正確格式：直接把 base64 放入 image_data
         data = {
             "model": MODEL,
             "messages": [
                 {
                     "role": "user",
                     "content": [
-                        {"type": "text", "text": f"![]({file_url}) Describe this image in detail."}
+                        {"type": "text", "text": "Describe this image in detail."},
+                        {"type": "image_data", "image_data": {"b64": image_b64}}
                     ]
                 }
             ]
@@ -60,17 +39,17 @@ def process(payload: dict):
         resp = requests.post(
             "https://router.huggingface.co/v1/chat/completions",
             headers={"Authorization": f"Bearer {HF_TOKEN}"},
-            data={"payload": json.dumps(data)},
+            json=data,
             timeout=60
         )
 
         if resp.status_code != 200:
-            return {"error": f"VLM API error: {resp.status_code}, {resp.text}"}
+            return {"error": f"HF VLM error: {resp.status_code}, {resp.text}"}
 
         try:
             vlm_text = resp.json()["choices"][0]["message"]["content"][0]["text"]
-        except (KeyError, IndexError, json.JSONDecodeError) as e:
-            return {"error": f"Failed to parse VLM response: {e}, Response text: {resp.text}"}
+        except Exception as e:
+            return {"error": f"Failed to parse VLM response: {e}, Response={resp.text}"}
 
         return {
             "received": True,