Meta-Eval/EvalHarness
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Read timeout 300→600s base (GLM gateway 30+s startup on 100k+ token inputs)

Browse Source 
Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
sora 2026-09-11 09:24:16 +00:00
parent 7ada0113ad
commit ff5d40aad5
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
evalharness/model/adapter.py
View File

@ -481,7 +481,9 @@ class OpenAICompatible(ModelAdapter):
# read timeout scales with the generation budget: a 32k-token
# CoT legitimately takes 10+ minutes; a fixed 300s timeout would
# kill and retry it forever (looks like a hang)
_rt = max(self.extra.get('timeout', 300),
# generous: GLM gateway takes 30+ seconds to start responding
# on 100k+ token inputs, even before any generation begins
_rt = max(self.extra.get('timeout', 600),
int(payload.get('max_tokens') or 0) * 0.15)
async with httpx.AsyncClient(timeout=httpx.Timeout(
connect=self.extra.get('connect_timeout', 15),