Add general:posttrain:dev suite

[finbarrtimbers]

Summary

• New general:posttrain:dev suite combining math:posttrain:dev, SciCode, IFBench, and GPQA Diamond
• Uses AVERAGE_OF_AVERAGES aggregation, matching sibling dev-style suites
• Expands to 12 underlying tasks

Test plan

• get_suite("general:posttrain:dev") resolves and expands correctly
• Pre-commit ruff format / check pass

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2ac3ed1d93

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Remove external eval from task suite definition

Replace "scicode" here with a real task spec (or split this into a separate external-eval workflow), because task suites are validated through validate_tasks() which only accepts entries where task_exists() or suite_exists() is true (src/olmo_eval/common/configs.py:72-101). scicode is registered as an external eval name (SciCodeExternalEval.name returns "scicode" in src/olmo_eval/evals/external/benchmarks/scicode/eval.py:79-82), not a task, so general:posttrain:dev expands to an invalid spec and cannot be executed via normal task/suite runs.

Useful? React with 👍 / 👎.