diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 6bfd2ac9dc..0c29cfd85e 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -2660,13 +2660,13 @@ kimik2.5-fp4-b200-vllm: - isl: 1024 osl: 1024 search-space: - - { tp: 8, ep: 1, conc-start: 4, conc-end: 4 } - - { tp: 4, ep: 1, conc-start: 4, conc-end: 64 } + - { tp: 8, ep: 1, conc-start: 1, conc-end: 4 } + - { tp: 4, ep: 1, conc-start: 1, conc-end: 128 } - isl: 8192 osl: 1024 search-space: - - { tp: 8, ep: 1, conc-start: 4, conc-end: 4 } - - { tp: 4, ep: 1, conc-start: 4, conc-end: 64 } + - { tp: 8, ep: 1, conc-start: 1, conc-end: 4 } + - { tp: 4, ep: 1, conc-start: 1, conc-end: 128 } # NOTE: At the time of submission, https://docs.vllm.ai/projects/recipes/en/latest/moonshotai/Kimi-K2.5.html # does not have a B300-specific recipe, so this config reuses the existing diff --git a/perf-changelog.yaml b/perf-changelog.yaml index fbe943b5f7..129eb5b2c8 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -3508,3 +3508,9 @@ - "Update vLLM image from v0.21.0 to v0.22.0" - "Expand concurrency sweep for the 1k/1k and 8k/1k cells: TP4/EP1 conc 4-64 -> 1-128, TP8/EP1 conc-start 4 -> 1 (conc-end 4 unchanged)." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1674 + +- config-keys: + - kimik2.5-fp4-b200-vllm + description: + - "Expand concurrency sweep for the 1k/1k and 8k/1k cells: TP4/EP1 conc 4-64 -> 1-128, TP8/EP1 conc-start 4 -> 1 (conc-end 4 unchanged)." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1677