mlcommons / mlcommons/inference
Request to reduce size of LLM checkpoints hosted by MLCommons
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.6k
- Forks
- 650
- Avg merge
- 1d 22h
- Merged PRs (30d)
- 6
Description
The current checkpoints hosted by MLCommons are large, and using them takes more storage space than necessary. As a rule-of-thumb, checkpoints require parameter-count * byte/parameter storage, so an 8B model with bf16 weights requires 16GB storage, 70B requires 140GB storage, and so on.
Contributor guide
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.
Research direction
Begin by reviewing how MLCommons hosts and consumes the current LLM checkpoints in the repository. Identify which checkpoints and storage formats are in scope, then clarify an acceptable reduction approach and validation criteria before attempting the change.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100