mlcommons / mlcommons/inference
LLaMA2-70B SUT_API.py batching logic
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.6k
- Forks
- 650
- Avg merge
- 1d 22h
- Merged PRs (30d)
- 6
Description
When I try to run the llama2-70b performance benchmark with a vLLM backend server, SUT_API.py seems to overwrite the batch size with the total sample count. Is this intentional or do you expect users to change this setting? Also has the Server mode been implemented for GPU runs yet?
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
Start with SUT_API.py in the LLaMA2-70B benchmark and trace how the batch size is assigned when using a vLLM backend server. Check the GPU Server mode path as well. Done means documenting or correcting the intended batching behavior and determining whether Server mode is implemented for GPU runs.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- machine-learning, python
- Domain
- machine-learning, performance
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100