danvics/haiku.rag
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Set explicit 120s timeout on the vllm reranker HTTP client

Browse source
This commit is contained in:
Yiorgis Gozadinos 2026-07-24 12:34:31 +03:00
parent 543aba7547
commit f31060e741
No known key found for this signature in database
1 changed files with 3 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
haiku_rag_slim/haiku/rag/reranking/vllm.py
View file

@ -28,7 +28,9 @@ class VLLMReranker(RerankerBase):
self._model = model self._model = model
self._base_url = base_url self._base_url = base_url
# One client reused across rerank calls (connection kept alive). # 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: async def aclose(self) -> None:
await self._client.aclose() await self._client.aclose()