Upload app.py

app.py

@@ -1,9 +1,10 @@
 import os
 import time
+import asyncio
 
 import gradio as gr
 from openai import OpenAI
-from vectra.solve import SolveConfig, solve
+from vectra.solve import SolveConfig, _solve_with_client
 
 DEFAULT_OPENROUTER_BASE = "https://openrouter.ai/api"
 DEFAULT_MODEL = "openai/gpt-4o-mini"
@@ -51,6 +52,49 @@ def _apply_vectra_env(api_key: str, vectra_base: str, model: str) -> None:
     os.environ["OPENAI_MODEL"] = model
 
 
+class OpenRouterVectraClient:
+    """Async chat adapter backed by the sync OpenAI SDK.
+
+    This avoids async client shutdown issues under certain Gradio event-loop contexts.
+    """
+
+    def __init__(self, *, api_key: str, sdk_base: str, model: str):
+        self._client = OpenAI(base_url=sdk_base, api_key=api_key)
+        self._model = model
+
+    async def chat(
+        self,
+        messages,
+        *,
+        model: str | None = None,
+        temperature: float = 0.7,
+        max_tokens: int | None = None,
+        n: int = 1,
+        extra: dict | None = None,
+    ) -> list[str]:
+        def _request() -> list[str]:
+            kwargs = {
+                "model": model or self._model,
+                "messages": [{"role": m.role, "content": m.content} for m in messages],
+                "temperature": float(temperature),
+                "n": int(n),
+            }
+            if max_tokens is not None:
+                kwargs["max_tokens"] = int(max_tokens)
+            if extra:
+                kwargs["extra_body"] = dict(extra)
+
+            response = self._client.chat.completions.create(**kwargs)
+            outputs: list[str] = []
+            for choice in response.choices:
+                outputs.append(_normalize_content(choice.message.content))
+            if not outputs:
+                raise RuntimeError("No choices returned from model")
+            return outputs
+
+        return await asyncio.to_thread(_request)
+
+
 def _normalize_content(content) -> str:
     if isinstance(content, str):
         return content.strip()
@@ -141,11 +185,12 @@ def _vectra_infer(
     max_calls: int,
     max_concurrency: int,
 ):
-    api_key, _, vectra_base, model = _resolve_openrouter_config(model_override)
+    api_key, sdk_base, vectra_base, model = _resolve_openrouter_config(model_override)
     _apply_vectra_env(api_key, vectra_base, model)
+    client = OpenRouterVectraClient(api_key=api_key, sdk_base=sdk_base, model=model)
 
     t0 = time.perf_counter()
-    result = solve(
+    result = _solve_with_client(
         prompt,
         SolveConfig(
             model=model,
@@ -155,6 +200,7 @@ def _vectra_infer(
             temperature=float(temperature),
             return_trace=True,
         ),
+        client=client,
     )
     elapsed = round(time.perf_counter() - t0, 3)