diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 5404ab235..54c960524 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1794,3 +1794,12 @@ - "DSv4-Pro on MI355X exceeded the 3h cap (STEP CANCELLED DUE TO TIME LIMIT) due to ~30min MoE JIT compile plus slow torch-fallback kernels (SGLANG_HACK_FLASHMLA_BACKEND=torch et al.) from sgl-project/sglang#23608" - "300 minutes matches the GH Actions outer timeout-minutes cap in benchmark-tmpl.yml" - "Retriggering dsv4-fp8-mi355x-sglang" + +- config-keys: + - dsv4-fp8-mi355x-sglang + description: + - "Bump MI355X SLURM allocation from --time=180 to --time=300 in runners/launch_mi355x-amds.sh" + - "DSv4-Pro on MI355X exceeded the 3h cap (STEP CANCELLED DUE TO TIME LIMIT) due to ~30min MoE JIT compile plus slow torch-fallback kernels (SGLANG_HACK_FLASHMLA_BACKEND=torch et al.) from sgl-project/sglang#23608" + - "300 minutes matches the GH Actions outer timeout-minutes cap in benchmark-tmpl.yml" + - "Retriggering dsv4-fp8-mi355x-sglang" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1148