mlcommons / mlcommons/training

Gres defaults to 8 regardless of config.sh

Open
#824 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
1.8k
Forks
594
PR merge metrics
No merged PRs in 30d

Description

The config has users specify the --gpus-per-node parameter however in the code it defaults to 8. This breaks on newer systems such as GB200 w/ 4 gpu's per node.

https://github.com/mlcommons/training/blob/751d033cea0d70f83ec2131d55d4373b1f0b52c5/large_language_model_pretraining/nemo/pretrain_llama31.py#L78

Error message is:

sbatch: error: Batch job submission failed: Requested node configuration is not available

I'll create a PR to update.

Contributor guide

Open the contributing guide

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.

Research direction

Start at large_language_model_pretraining/nemo/pretrain_llama31.py around line 78 and inspect how config.sh supplies --gpus-per-node. Reproduce the Slurm submission with a four-GPU-per-node configuration, then verify that Gres uses the configured value instead of defaulting to 8.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
2/5
Estimated time
1-3 hours
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
55/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.