diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index 7700edf09..dcd6753fa 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -534,7 +534,7 @@ minimaxm2.5-fp8-mi355x-vllm: - { tp: 8, ep: 8, conc-start: 2, conc-end: 2 } minimaxm2.5-fp8-mi355x-atom: - image: rocm/atom:rocm7.2.1-ubuntu24.04-pytorch2.9.1-atom0.1.2 + image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: mi355x @@ -545,15 +545,13 @@ minimaxm2.5-fp8-mi355x-atom: - isl: 1024 osl: 1024 search-space: - - { tp: 2, conc-start: 4, conc-end: 128 } - - { tp: 4, conc-start: 4, conc-end: 128 } - - { tp: 8, ep: 8, conc-start: 32, conc-end: 256 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } - isl: 8192 osl: 1024 search-space: - - { tp: 2, conc-start: 4, conc-end: 128 } - - { tp: 4, conc-start: 4, conc-end: 128 } - - { tp: 8, ep: 8, conc-start: 32, conc-end: 256 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } minimaxm2.5-fp4-mi355x-vllm: image: vllm/vllm-openai-rocm:v0.19.1 diff --git a/benchmarks/single_node/minimaxm2.5_fp8_mi355x_atom.sh b/benchmarks/single_node/minimaxm2.5_fp8_mi355x_atom.sh index ca84f8228..2a8c67da0 100755 --- a/benchmarks/single_node/minimaxm2.5_fp8_mi355x_atom.sh +++ b/benchmarks/single_node/minimaxm2.5_fp8_mi355x_atom.sh @@ -39,6 +39,7 @@ fi # Start GPU monitoring (power, temperature, clocks every second) start_gpu_monitor +MEM_FRAC_STATIC=0.9 set -x @@ -47,6 +48,7 @@ python3 -m atom.entrypoints.openai_server \ --server-port $PORT \ -tp $TP \ --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \ + --gpu-memory-utilization $MEM_FRAC_STATIC \ --trust-remote-code \ > $SERVER_LOG 2>&1 & diff --git a/perf-changelog.yaml b/perf-changelog.yaml index a6c811748..f3fbf6bd9 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1812,3 +1812,10 @@ - "Topologies: low-conc 1p1d-dep8-tep8 (4 nodes, mirrored from NVIDIA srt-slurm PR #71 with offload kept and numa-bind dropped); mid 1p1d-dep8-dep16 (6 nodes) and high 3p1d-dep8-dep16 (10 nodes) hand-rolled, structurally derived from the kimi-k2.5 1k/1k pattern" - "Recipes stored under benchmarks/multi_node/srt-slurm-recipes/ and overlaid onto the upstream srt-slurm checkout at runtime" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1129 + +- config-keys: + - minimaxm2.5-fp8-mi355x-atom + description: + - "Update MiniMax-M2.5 FP8 MI355X Atom benchmark (rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post)" + - "Concurrency up to 256 on TP2/TP4" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1194