router eval: fixed 60 s embedding request timeout fails full runs on slower machines

Open Beginner friendly
#459 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
2/5
Estimated time
1-3 hours
Newbie friendliness
67/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
python
Domain
ai, backend

Research direction

Start with systems/superintelligence/scripts/meta_select.py and inspect _embed_at, then trace how evals/routing/score.py --mode auto calls it. The issue offers two possible fixes, so confirm the intended approach with the maintainers before changing the timeout or retry behavior. Done when full router evaluations tolerate slower embedding requests while preserving the refusal to write a keyword score as the router score.

Written by the indexing model from the issue text.

Description

enhancement

What happens

systems/superintelligence/scripts/meta_select.py sends every embedding request with a fixed 60 second timeout (_embed_at). A full router eval (evals/routing/score.py --mode auto) sends one request per prompt with the prompt plus the team profiles. On a slower or busy machine one of those requests can take longer than 60 seconds, and the whole run stops with:

embedding server failed during scoring at <url>: TimeoutError: timed out
refusing to write a keyword score as the router top-1

Re-running the same command on a quieter machine passes, so the result depends on machine load rather than the router.

Suggested fix

Either of these would do:

  • read the timeout from an environment variable (for example COCO_EMBED_TIMEOUT, default 60) so slower machines can raise it;
  • add a solo/serial mode to score.py that retries a timed-out request once before failing the run.

The refusal to write a keyword score as the router score should stay as it is.

Dominant language
HTML
Stars
347
Forks
12
Avg merge
2d 16h
Merged PRs (30d)
95

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from coco-research/coco

All issues in coco-research/coco

Similar issues

More AI Infra & Agents issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.