mlcommons / mlcommons/inference
Metrics definition and settings used for LLM benchmarks
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.6k
- Forks
- 650
- Avg merge
- 1d 22h
- Merged PRs (30d)
- 6
Description
Hello,
I have query regarding metrics definition and settings used in LLM benchmarks in MLCommons (https://mlcommons.org/benchmarks/inference-datacenter/).
For LLM-Q/A task in the benchmark table in above link:
- How TPOT and TTFT are calculated? Can you source code for it? For TPOT how many tokens are generated?
- For openocra dataset, input prompt provided is "system_prompt" + "question" or "question" only? Since openocra dataset has both colmuns
- For quality metrics in table the values for ROUGE-1 is its precision , recall or fmeasure?
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
The issue asks for definitions and source references for TPOT, TTFT, generated-token counts, OpenOrca prompt construction, and the ROUGE-1 value. Start by locating the benchmark metric definitions and the code handling the OpenOrca dataset; done means documenting clear answers with relevant source links, but no file or test is named.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- documentation, machine-learning
- Issue type
- Documentation
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100