From f31060e7417597abae3e87081fa9d46b168e1b31 Mon Sep 17 00:00:00 2001 From: Yiorgis Gozadinos Date: Fri, 24 Jul 2026 12:34:31 +0300 Subject: [PATCH] Set explicit 120s timeout on the vllm reranker HTTP client --- haiku_rag_slim/haiku/rag/reranking/vllm.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/haiku_rag_slim/haiku/rag/reranking/vllm.py b/haiku_rag_slim/haiku/rag/reranking/vllm.py index 4a5e4d22..a2b40087 100644 --- a/haiku_rag_slim/haiku/rag/reranking/vllm.py +++ b/haiku_rag_slim/haiku/rag/reranking/vllm.py @@ -28,7 +28,9 @@ class VLLMReranker(RerankerBase): self._model = model self._base_url = base_url # One client reused across rerank calls (connection kept alive). - self._client = httpx.AsyncClient() + # Multimodal document batches can take far longer than httpx's 5s + # default timeout to score. + self._client = httpx.AsyncClient(timeout=httpx.Timeout(120.0)) async def aclose(self) -> None: await self._client.aclose()