OpenEuroLLM / OpenEuroLLM/Taskboard
[Baby 9B dense] Downstream evals
Open
@jindrahelcl is already working on this.
Since Jul 24, 2026.
T4.4 - model training
T5.1 - static evals
- Dominant language
- No language data
- Stars
- 3
- Forks
- 0
- PR merge metrics
- No merged PRs in 30d
Description
Goal: Evaluate baby (aka prelude aka oellm-9b) intermediate checkpoints on downstream task evals.
Description: We are currently evaluating the checkpoints for every 0.5T tokens starting from 1T on the following task groups (in progress):
- dclm-core-22
- oellm-multilingual
- reasoning
Deliverable scope:
- Downstream eval numbers for intermediate as well as the final cooled down checkpoint on 10T tokens.
- Compare them with some baseline models of equivalent size: datamix 9B-80-20, apertus 8B, Olmo3 7B, EuroLLM 9B, Marin 8B, and final models of Qwen 3 8B, Qwen3.5-9B. See ellamind base-eval plot.
Dependencies:
- Checkpoint conversion from MegatronLM to HF format, current converted checkpoints can be found on HF.
- Some tasks in reasoning depend on
evalchemywhich are currently failing in container - For some tasks we need a chat template
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.