Update app.py

app.py

@@ -165,23 +165,30 @@ def analyze(req: VisionRequest):
 @app.post("/chat")
 def chat(req: ChatRequest):
 
-    image = decode_image(req.image) if req.image else None
-
     messages = req.messages
 
+    image = None
+    if req.image:
+        image = decode_image(req.image)
+
+    # ───── inject image safely ─────
     if image:
         messages[-1]["content"].insert(0, {
             "type": "image",
             "image": image
         })
 
+    # ───── IMPORTANT FIX HERE ─────
     inputs = processor.apply_chat_template(
         messages,
         add_generation_prompt=True,
+        tokenize=True,        # 🔴 THIS FIXES IT
         return_tensors="pt"
     )
 
-    inputs = inputs.to(model.device)
+    # now inputs is a tensor dict (NOT string anymore)
+
+    inputs = {k: v.to(model.device) for k, v in inputs.items()}
 
     with torch.no_grad():
         output = model.generate(