fix: ssr_mode=False + messages-format history (Gradio 6 runtime error)

app.py

@@ -92,7 +92,7 @@ else:
 
 def respond(
     message: str,
-    history: list[tuple[str, str]],
+    history: list[dict[str, str]],
     model: str,
     system_prompt: str,
     temperature: float,
@@ -116,14 +116,17 @@ def respond(
 
     is_first_response = not (history or [])
 
+    # Gradio 5+ ChatInterface(type="messages") gives history as
+    # [{"role": "user"|"assistant", "content": str}, ...] — same shape the
+    # OpenAI SDK wants, so we can pass it through with light filtering.
     messages: list[dict[str, str]] = []
     if system_prompt.strip():
         messages.append({"role": "system", "content": system_prompt.strip()})
-    for user_msg, assistant_msg in history or []:
-        if user_msg:
-            messages.append({"role": "user", "content": user_msg})
-        if assistant_msg:
-            messages.append({"role": "assistant", "content": assistant_msg})
+    for turn in history or []:
+        role = turn.get("role") if isinstance(turn, dict) else None
+        content = turn.get("content") if isinstance(turn, dict) else None
+        if role in ("user", "assistant") and content:
+            messages.append({"role": role, "content": content})
     messages.append({"role": "user", "content": message})
 
     try:
@@ -200,6 +203,7 @@ with gr.Blocks(title="QuickSilver Pro Chat") as demo:
             # keeps this compatible with both 5.x and 6.x.
             gr.ChatInterface(
                 fn=respond,
+                type="messages",
                 additional_inputs=[model_dropdown, system_prompt, temperature, max_tokens],
                 examples=[
                     ["Write a concise git commit message for: fixed off-by-one error in pagination"],
@@ -214,4 +218,9 @@ with gr.Blocks(title="QuickSilver Pro Chat") as demo:
 
 
 if __name__ == "__main__":
-    demo.queue(default_concurrency_limit=4, max_size=64).launch()
+    # ssr_mode=False — Gradio 6.x ships an experimental SSR path that runs a
+    # localhost reachability check at launch. Inside HF Spaces' container that
+    # check fails ("When localhost is not accessible, a shareable link must
+    # be created."), crashing the Space with RUNTIME_ERROR. Disabling SSR
+    # restores classic CSR rendering, which is what the Space ran on before.
+    demo.queue(default_concurrency_limit=4, max_size=64).launch(ssr_mode=False)