OpenEuroLLM / OpenEuroLLM/Taskboard
[32B Dense] Babysitting instructions
@tvosch is already working on this.
Since Jul 1, 2026.
- Dominant language
- No language data
- Stars
- 3
- Forks
- 0
- PR merge metrics
- No merged PRs in 30d
Description
Babysitting instructions for 32B model
Software setup
- Get access to Jupiter and the corresponding Strategic allocation
e-sta-openeurollm, Extreme Scale budgete-ext-2025e02-108and datasetsdatasets - Clone oellm-autoexp:
git clone https://github.com/OpenEuroLLM/oellm-autoexp.git --recurse-submodules - Checkout to pretraining branch:
git checkout prod/oellm_32b_dense - If no uv is installed, follow: https://github.com/OpenEuroLLM/oellm-autoexp#quickstart-using-pre-configured-enviroment-and-default-values
- Run a test job that does not interfere with the production job:
uv run python scripts/run_autoexp.py --config-name experiments/... - Check if you have permissions to write files:
getfacl /e/project1/e-sta-openeurollm/production_training/
getfacl /e/project1/e-sta-openeurollm/pre_production_training/
getfacl /e/scratch/e-sta-openeurollm/production_training/
getfacl /e/scratch/e-sta-openeurollm/pre_production_training/
- Set up W&B in .venv:
uv pip install wandbthenwandb login
Start a new training job
- Pull latest changes from
prod/oellm_32b_dense - Run in oellm-exp root:
uv run python scripts/run_autoexp.py --config-name experiments/oellm_32b_dense/oellm_32b_dense_production
All necessary files including logs, checkpoints, data, wandb/tensorboard are stored at path:
/e/project1/e-sta-openeurollm/production_training/oellm_32b_dense_prod_gbs4096_lr3e-4
To exclude nodes, add them to the autoexclusion file at
/e/project1/e-sta-openeurollm/production_training/jupiter_exclude_nodes.txt
[!NOTE]
In case you will start the new training job with auto_restart, this has to be performed over tmux (persistent connection from your local to login node of Leonardo) as the daemon spawning new jobs will work on a live terminal
Babysitting duties
Schedule here https://docs.google.com/spreadsheets/d/1uYidr_VgjRBXxdYZ4b-8UXQmDJFv_jpCMtsDF_Br5Ew/edit?gid=2087395347#gid=2087395347
-
Check running jobs:
... -
Monitor W&B for loss spikes and throughput (tokens/s/GPU) spikes at: https://wandb.ai/openeurollm-project/...
-
Automatic W&B sync should happen already: https://github.com/OpenEuroLLM/Taskboard/issues/340
-
Monitoring tool: https://github.com/OpenEuroLLM/oellm-monitoring
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.