Update app.py

app.py

@@ -77,7 +77,7 @@ body.light{
   overflow:hidden;
 }
 
-/* Bubble styling (best-effort: Gradio DOM can vary by version) */
+/* Best-effort bubble styling (DOM varies by Gradio version) */
 .chat_shell .message{
   border-radius:16px !important;
   border:1px solid var(--border) !important;
@@ -152,10 +152,7 @@ def get_headers():
     key = os.environ.get("HF_API_KEY")
     if not key:
         return None, "HF_API_KEY missing (Space Settings → Variables)"
-    return {
-        "Authorization": f"Bearer {key}",
-        "Content-Type": "application/json",
-    }, None
+    return {"Authorization": f"Bearer {key}", "Content-Type": "application/json"}, None
 
 def list_models(headers):
     r = requests.get(f"{BASE_URL}/models", headers=headers, timeout=20)
@@ -163,7 +160,9 @@ def list_models(headers):
         return None, f"HTTP {r.status_code}: {r.text}"
     data = r.json()
     models = [m.get("id") for m in data.get("data", []) if m.get("id")]
-    return models, None if models else (None, "No models returned")
+    if not models:
+        return None, "No models returned"
+    return models, None
 
 def chat_call(headers, model, messages, temperature, max_tokens):
     payload = {
@@ -188,30 +187,47 @@ def chat_call(headers, model, messages, temperature, max_tokens):
         return f"Unexpected response shape: {data}"
 
 # ----------------------------
-# Conversation management
+# History conversion utilities
 # ----------------------------
-MAX_TURNS = 14  # number of (user+assistant) pairs to send each request
-
-def trim_history(history):
-    """
-    history is Gradio Chatbot type='messages':
-      [{"role":"user","content":"..."}, {"role":"assistant","content":"..."}, ...]
-    """
-    if not history:
+MAX_TURNS = 14  # keep last N user+assistant pairs
+
+def ui_pairs_from_messages(msgs):
+    """Convert OpenAI messages -> Chatbot tuple list [(user, assistant), ...]"""
+    pairs = []
+    pending_user = None
+    for m in msgs or []:
+        role = m.get("role")
+        content = m.get("content", "")
+        if role == "user":
+            pending_user = content
+        elif role == "assistant":
+            if pending_user is None:
+                # assistant without user, show it anyway
+                pairs.append(("", content))
+            else:
+                pairs.append((pending_user, content))
+                pending_user = None
+    if pending_user is not None:
+        pairs.append((pending_user, ""))  # user message without assistant yet
+    return pairs
+
+def trimmed_messages(msgs):
+    """Trim to last MAX_TURNS pairs (2 messages each)."""
+    if not msgs:
         return []
-    return history[-MAX_TURNS * 2 :]
+    return msgs[-MAX_TURNS * 2 :]
 
-def build_messages(system_prompt, history, user_msg):
+def build_messages(system_prompt, msgs, user_msg):
     return (
         [{"role": "system", "content": system_prompt}]
-        + trim_history(history)
+        + trimmed_messages(msgs)
         + [{"role": "user", "content": user_msg}]
     )
 
 # ----------------------------
 # Gradio app
 # ----------------------------
-with gr.Blocks(title="Chat", css=CSS) as demo:
+with gr.Blocks(title="Chat") as demo:
     with gr.Column(elem_id="wrap"):
         with gr.Row(elem_classes=["header"]):
             gr.Markdown("## My AI Chatbot")
@@ -223,21 +239,23 @@ with gr.Blocks(title="Chat", css=CSS) as demo:
             with gr.Row():
                 model_dd = gr.Dropdown(label="Model (from /v1/models)", choices=[], value=None, interactive=True)
                 refresh_btn = gr.Button("Refresh models", elem_classes=["secondary"])
+
             model_override = gr.Textbox(
                 label="Model override (optional)",
-                placeholder="If dropdown models fail, paste a chat-capable model id here (takes precedence).",
-            )
-            system_prompt = gr.Textbox(
-                label="System prompt",
-                value="You are a helpful assistant.",
+                placeholder="Paste a chat-capable model id here (takes precedence over dropdown).",
             )
+
+            system_prompt = gr.Textbox(label="System prompt", value="You are a helpful assistant.")
             with gr.Row():
                 temperature = gr.Slider(0.0, 1.5, value=0.7, step=0.05, label="Temperature")
                 max_tokens = gr.Slider(64, 2048, value=800, step=32, label="Max tokens")
 
         with gr.Column(elem_classes=["chat_shell"]):
-            # CRITICAL: type="messages" ensures history is OpenAI-style dicts
-            chatbot = gr.Chatbot(type="messages", height=560)
+            # IMPORTANT: no `type=` here (your Gradio doesn't support it)
+            chatbot = gr.Chatbot(height=560)
+
+        # Internal message state (OpenAI-style dict list)
+        msg_state = gr.State([])
 
         with gr.Column(elem_classes=["composer"]):
             msg = gr.Textbox(placeholder="Message…", show_label=False, lines=1)
@@ -249,62 +267,64 @@ with gr.Blocks(title="Chat", css=CSS) as demo:
         headers, err = get_headers()
         if err:
             return f"❌ {err}", gr.update(choices=[], value=None)
-
         models, err = list_models(headers)
-        if err or not models:
-            return f"❌ {err or 'No models found'}", gr.update(choices=[], value=None)
-
-        # do NOT assume first model is chat-capable; just pick first as a convenience
+        if err:
+            return f"❌ {err}", gr.update(choices=[], value=None)
         return f"✅ Models loaded ({len(models)}).", gr.update(choices=models, value=models[0])
 
     def init():
         return do_refresh_models()
 
-    def respond(user_msg, history, model_dd_value, model_text, sys_prompt, temp, mx):
-        history = history or []
+    def respond(user_msg, msgs, model_dd_value, model_text, sys_prompt, temp, mx):
+        msgs = msgs or []
         user_msg = (user_msg or "").strip()
         if not user_msg:
-            return "", history
+            return "", ui_pairs_from_messages(msgs), msgs
 
         headers, err = get_headers()
         if err:
-            history = history + [
+            msgs = msgs + [
                 {"role": "user", "content": user_msg},
                 {"role": "assistant", "content": f"Setup error: {err}"},
             ]
-            return "", history
+            return "", ui_pairs_from_messages(msgs), msgs
 
-        # Prefer override, otherwise dropdown
         model = (model_text or "").strip() or (model_dd_value or "").strip()
         if not model:
-            history = history + [
+            msgs = msgs + [
                 {"role": "user", "content": user_msg},
-                {"role": "assistant", "content": "No model selected. Choose one from the dropdown or set Model override."},
+                {"role": "assistant", "content": "No model selected. Choose one or set Model override."},
             ]
-            return "", history
+            return "", ui_pairs_from_messages(msgs), msgs
 
-        messages = build_messages(sys_prompt or "You are a helpful assistant.", history, user_msg)
-        bot = chat_call(headers, model, messages, temp, mx)
+        # Build request messages safely
+        req_messages = build_messages(sys_prompt or "You are a helpful assistant.", msgs, user_msg)
+        bot = chat_call(headers, model, req_messages, temp, mx)
 
-        history = history + [
+        msgs = msgs + [
             {"role": "user", "content": user_msg},
             {"role": "assistant", "content": bot},
         ]
-        return "", history
+
+        return "", ui_pairs_from_messages(msgs), msgs
+
+    def do_clear():
+        return [], []
 
     demo.load(init, outputs=[status, model_dd])
     refresh_btn.click(do_refresh_models, outputs=[status, model_dd])
 
     send.click(
         respond,
-        inputs=[msg, chatbot, model_dd, model_override, system_prompt, temperature, max_tokens],
-        outputs=[msg, chatbot],
+        inputs=[msg, msg_state, model_dd, model_override, system_prompt, temperature, max_tokens],
+        outputs=[msg, chatbot, msg_state],
     )
     msg.submit(
         respond,
-        inputs=[msg, chatbot, model_dd, model_override, system_prompt, temperature, max_tokens],
-        outputs=[msg, chatbot],
+        inputs=[msg, msg_state, model_dd, model_override, system_prompt, temperature, max_tokens],
+        outputs=[msg, chatbot, msg_state],
     )
-    clear.click(lambda: [], outputs=chatbot)
+    clear.click(do_clear, outputs=[chatbot, msg_state])
 
-demo.launch()
+# Gradio 6.x: pass css to launch(), not Blocks()
+demo.launch(css=CSS)