For each SKU,
- Search vLLM release notes / commit history to check what changes must be made to recipes
- Search vLLM cookbook recipes to see if any updates have been made
- Perform e2e test confirming the image works and perf is stable
- Update perf-changelog.yaml accordingly
Consider the current version for each SKU. For instance, MI355X is on a very old (not even upstream) version of vLLM and therefore you must do some deeper digging into all changes SINCE that release.
While the cookbook recipe (https://github.com/vllm-project/recipes/blob/main/OpenAI/GPT-OSS.md) is a good starting point reference for all the SKUs, you must manually look through release notes and release commits since the most recently used version and see what flags / CLI args must change in the benchmarks/ recipes to achieve optimal perf.
For each SKU,
Consider the current version for each SKU. For instance, MI355X is on a very old (not even upstream) version of vLLM and therefore you must do some deeper digging into all changes SINCE that release.
While the cookbook recipe (https://github.com/vllm-project/recipes/blob/main/OpenAI/GPT-OSS.md) is a good starting point reference for all the SKUs, you must manually look through release notes and release commits since the most recently used version and see what flags / CLI args must change in the
benchmarks/recipes to achieve optimal perf.