Update app.py

app.py

@@ -1,5 +1,3 @@
-is this code correct??
-
 import gradio as gr
 from transformers import pipeline
 from PIL import Image
@@ -18,40 +16,32 @@ pipe = pipeline("image-text-to-text", model="Qwen/Qwen2-VL-2B-Instruct")
 
 # ---------- robust extractor: returns ONLY the model text ----------
 def _only_model_text(out) -> str:
-    # Case A: chat transcript list → grab the last assistant turn
+    # Case 1: pipelines often return {"generated_text": "..."}
+    if isinstance(out, dict) and "generated_text" in out:
+        return out["generated_text"]
+
+    # Case 2: list of dicts (mixed roles)
     if isinstance(out, list):
-        # Prefer any assistant turn from the end
-        for item in reversed(out):
+        # Prefer any dict with generated_text first
+        for item in out:
+            if isinstance(item, dict) and "generated_text" in item:
+                return item["generated_text"]
+        # Otherwise find assistant role
+        for item in out:
             if isinstance(item, dict) and item.get("role") == "assistant":
                 content = item.get("content")
                 if isinstance(content, str):
                     return content
                 if isinstance(content, list):
-                    # collect only text segments, ignore images
-                    texts = []
-                    for seg in content:
-                        if isinstance(seg, dict) and seg.get("type") == "text":
-                            t = seg.get("text")
-                            if isinstance(t, str):
-                                texts.append(t)
-                    if texts:
-                        return "\n".join(texts)
-        # Also handle [{'generated_text': '...'}] pattern if present
-        for item in out:
-            if isinstance(item, dict) and "generated_text" in item:
-                return item["generated_text"]
-
-    # Case B: dict with generated_text
-    if isinstance(out, dict) and "generated_text" in out:
-        return out["generated_text"]
-
-    # Fallback: stringify safely
-    try:
-        import json as _json
-        return _json.dumps(out, ensure_ascii=False)
-    except Exception:
-        return str(out)
-
+                    # collect text pieces within the assistant content
+                    chunks = []
+                    for c in content:
+                        if isinstance(c, dict) and c.get("type") == "text" and isinstance(c.get("text"), str):
+                            chunks.append(c["text"])
+                    if chunks:
+                        return "\n".join(chunks)
+    # Fallback
+    return str(out)
 
 def infer(file_obj, prompt):
     if file_obj is None:
@@ -86,10 +76,8 @@ def infer(file_obj, prompt):
     out = pipe(text=messages, max_new_tokens=256)
 
     # return ONLY the assistant text
-    out = pipe(text=messages, max_new_tokens=256)
     return _only_model_text(out)
 
-
 # ---------- Gradio UI ----------
 with gr.Blocks(
     title="Qwen2-VL-2B — File + Prompt",
@@ -105,9 +93,14 @@ with gr.Blocks(
     run_btn = gr.Button("Run")
 
     # output textbox that expands (via CSS above)
-    resp_out = gr.Markdown(elem_id="resp_out")
+    resp_out = gr.Textbox(
+        label="Model Response",
+        lines=8,
+        show_copy_button=True,
+        elem_id="resp_out"
+    )
 
     run_btn.click(fn=infer, inputs=[file_in, prompt_in], outputs=[resp_out])
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()