From 3ed70d614f5b962846b5842d1c6bd1eb5fb7f658 Mon Sep 17 00:00:00 2001 From: seungrokj Date: Thu, 16 Apr 2026 08:22:47 +0000 Subject: [PATCH 1/8] atom minimax fp4 on mi355x Signed-off-by: seungrokj --- .github/configs/amd-master.yaml | 26 ++++++ .../minimaxm2.5_fp4_mi355x_atom.sh | 80 +++++++++++++++++++ 2 files changed, 106 insertions(+) create mode 100644 benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index 89318004b..3c1a8ee8e 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -415,6 +415,32 @@ minimaxm2.5-fp8-mi355x-atom: - { tp: 4, conc-start: 4, conc-end: 128 } - { tp: 8, ep: 8, conc-start: 32, conc-end: 256 } +minimaxm2.5-fp4-mi355x-atom: + # TODO: + image: TBD + model: MiniMaxAI/MiniMax-M2.5 + model-prefix: minimaxm2.5 + runner: mi355x + precision: fp4 + framework: atom + multinode: false + seq-len-configs: + - isl: 1024 + osl: 1024 + # TODO: + search-space: + - { tp: 1, conc-start: 4, conc-end: 256 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 8, conc-start: 4, conc-end: 256 } + - isl: 8192 + osl: 1024 + search-space: + - { tp: 1, conc-start: 4, conc-end: 256 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 8, conc-start: 4, conc-end: 256 } + minimaxm2.5-fp8-mi300x-vllm: image: vllm/vllm-openai-rocm:v0.16.0 model: MiniMaxAI/MiniMax-M2.5 diff --git a/benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh b/benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh new file mode 100644 index 000000000..ca84f8228 --- /dev/null +++ b/benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh @@ -0,0 +1,80 @@ +#!/usr/bin/env bash + +source "$(dirname "$0")/../benchmark_lib.sh" + +check_env_vars \ + MODEL \ + TP \ + CONC \ + ISL \ + OSL \ + RANDOM_RANGE_RATIO \ + RESULT_FILENAME \ + EP_SIZE \ + DP_ATTENTION + +if [[ -n "$SLURM_JOB_ID" ]]; then + echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME" +fi + +echo "TP: $TP, CONC: $CONC, ISL: $ISL, OSL: $OSL, EP_SIZE: $EP_SIZE, DP_ATTENTION: $DP_ATTENTION" + +SERVER_LOG=/workspace/server.log +PORT=${PORT:-8888} + +export OMP_NUM_THREADS=1 + +# Calculate max-model-len based on ISL and OSL +if [ "$ISL" = "1024" ] && [ "$OSL" = "1024" ]; then + CALCULATED_MAX_MODEL_LEN="" +else + CALCULATED_MAX_MODEL_LEN=" --max-model-len 10240 " +fi + +if [ "$EP_SIZE" -gt 1 ]; then + EP=" --enable-expert-parallel" +else + EP=" " +fi + +# Start GPU monitoring (power, temperature, clocks every second) +start_gpu_monitor + +set -x + +python3 -m atom.entrypoints.openai_server \ + --model $MODEL \ + --server-port $PORT \ + -tp $TP \ + --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \ + --trust-remote-code \ + > $SERVER_LOG 2>&1 & + +SERVER_PID=$! + +# Wait for server to be ready +wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID" + +export PYTHONDONTWRITEBYTECODE=1 +run_benchmark_serving \ + --model "$MODEL" \ + --port "$PORT" \ + --backend vllm \ + --input-len "$ISL" \ + --output-len "$OSL" \ + --random-range-ratio "$RANDOM_RANGE_RATIO" \ + --num-prompts "$((CONC * 10))" \ + --max-concurrency "$CONC" \ + --result-filename "$RESULT_FILENAME" \ + --result-dir /workspace/ \ + --trust-remote-code + +# After throughput, run evaluation only if RUN_EVAL is true +if [ "${RUN_EVAL}" = "true" ]; then + run_eval --framework lm-eval --port "$PORT" + append_lm_eval_summary +fi + +# Stop GPU monitoring +stop_gpu_monitor +set +x From ecce815b61c541a5eecbc073876eb884e0398d3c Mon Sep 17 00:00:00 2001 From: seungrokj <144636725+seungrokj@users.noreply.github.com> Date: Fri, 17 Apr 2026 17:43:29 +0900 Subject: [PATCH 2/8] Update amd-master.yaml --- .github/configs/amd-master.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index 3c1a8ee8e..21010f688 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -418,7 +418,7 @@ minimaxm2.5-fp8-mi355x-atom: minimaxm2.5-fp4-mi355x-atom: # TODO: image: TBD - model: MiniMaxAI/MiniMax-M2.5 + model: amd/MiniMax-M2.5-MXFP4 model-prefix: minimaxm2.5 runner: mi355x precision: fp4 From d2e4cb5c47bb2cc093579dacea02f061e43c2c86 Mon Sep 17 00:00:00 2001 From: seungrokj Date: Sat, 25 Apr 2026 14:10:29 +0900 Subject: [PATCH 3/8] Update minimaxm2.5-fp4-mi355x-atom config and add minimaxm2.5-fp4-mi355x-vllm Co-Authored-By: Claude Sonnet 4.6 --- .github/configs/amd-master.yaml | 42 ++++++++++++++++++++++++--------- 1 file changed, 31 insertions(+), 11 deletions(-) diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index 21010f688..718579b6d 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -416,8 +416,7 @@ minimaxm2.5-fp8-mi355x-atom: - { tp: 8, ep: 8, conc-start: 32, conc-end: 256 } minimaxm2.5-fp4-mi355x-atom: - # TODO: - image: TBD + image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post model: amd/MiniMax-M2.5-MXFP4 model-prefix: minimaxm2.5 runner: mi355x @@ -427,19 +426,40 @@ minimaxm2.5-fp4-mi355x-atom: seq-len-configs: - isl: 1024 osl: 1024 - # TODO: search-space: - - { tp: 1, conc-start: 4, conc-end: 256 } - - { tp: 2, conc-start: 4, conc-end: 256 } - - { tp: 4, conc-start: 4, conc-end: 256 } - - { tp: 8, conc-start: 4, conc-end: 256 } + - { tp: 1, conc-start: 4, conc-end: 1024 } + - { tp: 2, conc-start: 4, conc-end: 1024 } + - { tp: 4, conc-start: 4, conc-end: 128 } + - { tp: 8, conc-start: 4, conc-end: 16 } - isl: 8192 osl: 1024 search-space: - - { tp: 1, conc-start: 4, conc-end: 256 } - - { tp: 2, conc-start: 4, conc-end: 256 } - - { tp: 4, conc-start: 4, conc-end: 256 } - - { tp: 8, conc-start: 4, conc-end: 256 } + - { tp: 1, conc-start: 4, conc-end: 1024 } + - { tp: 2, conc-start: 4, conc-end: 1024 } + - { tp: 4, conc-start: 4, conc-end: 128 } + - { tp: 8, conc-start: 4, conc-end: 16 } + +minimaxm2.5-fp4-mi355x-vllm: + image: vllm/vllm-openai-rocm:v0.19.1 + model: amd/MiniMax-M2.5-MXFP4 + model-prefix: minimaxm2.5 + runner: mi355x + precision: fp4 + framework: vllm + multinode: false + seq-len-configs: + - isl: 1024 + osl: 1024 + search-space: + - { tp: 1, conc-start: 4, conc-end: 32 } + - { tp: 2, conc-start: 4, conc-end: 64 } + - { tp: 4, conc-start: 4, conc-end: 64 } + - isl: 8192 + osl: 1024 + search-space: + - { tp: 1, conc-start: 4, conc-end: 32 } + - { tp: 2, conc-start: 4, conc-end: 64 } + - { tp: 4, conc-start: 4, conc-end: 64 } minimaxm2.5-fp8-mi300x-vllm: image: vllm/vllm-openai-rocm:v0.16.0 From d6131383916a5f8501bd1655bda64bafa5619a4b Mon Sep 17 00:00:00 2001 From: seungrokj Date: Sat, 25 Apr 2026 14:13:46 +0900 Subject: [PATCH 4/8] Fill in perf-changelog description for minimaxm2.5-fp4-mi355x-atom Co-Authored-By: Claude Sonnet 4.6 --- perf-changelog.yaml | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index a6c811748..478df6692 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-fp4-mi355x-atom + description: + - "Add MiniMax-M2.5 MXFP4 MI355X Atom benchmark (rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post)" + - "Single-node sweep: TP1–TP8, 1k/1k and 8k/1k ISL/OSL" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1042 From 0ba47da7d1f0310255fba03347078a05daf7cdbb Mon Sep 17 00:00:00 2001 From: seungrokj Date: Mon, 27 Apr 2026 21:00:36 +0900 Subject: [PATCH 5/8] Update minimaxm2.5-fp8-mi355x-atom config and add perf-changelog entry Co-Authored-By: Claude Sonnet 4.6 --- .github/configs/amd-master.yaml | 12 +++++------- perf-changelog.yaml | 6 ++++++ 2 files changed, 11 insertions(+), 7 deletions(-) diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index cec13ccb9..d35086dcc 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-atom: image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 478df6692..981cf6d10 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1819,3 +1819,9 @@ - "Add MiniMax-M2.5 MXFP4 MI355X Atom benchmark (rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post)" - "Single-node sweep: TP1–TP8, 1k/1k and 8k/1k ISL/OSL" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1042 + +- config-keys: + - minimaxm2.5-fp8-mi355x-atom + description: + - "" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/ \ No newline at end of file From ebc642480bc833d2089be49560936b1c0df8f63b Mon Sep 17 00:00:00 2001 From: seungrokj Date: Mon, 27 Apr 2026 21:03:18 +0900 Subject: [PATCH 6/8] Fill in PR link for minimaxm2.5-fp8-mi355x-atom perf-changelog entry Co-Authored-By: Claude Sonnet 4.6 --- perf-changelog.yaml | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 981cf6d10..61a7d0b61 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1823,5 +1823,6 @@ - config-keys: - minimaxm2.5-fp8-mi355x-atom description: - - "" - pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/ \ No newline at end of file + - "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 \ No newline at end of file From 9b0c6c5bf87042b4ad1eebc4a07b0c98d724b17d Mon Sep 17 00:00:00 2001 From: seungrokj Date: Mon, 27 Apr 2026 21:15:05 +0900 Subject: [PATCH 7/8] Update minimaxm2.5-fp8-mi355x-atom: add gpu-memory-utilization, remove fp4 atom config Co-Authored-By: Claude Sonnet 4.6 --- .github/configs/amd-master.yaml | 24 ------ .../minimaxm2.5_fp4_mi355x_atom.sh | 80 ------------------- .../minimaxm2.5_fp8_mi355x_atom.sh | 2 + perf-changelog.yaml | 7 -- 4 files changed, 2 insertions(+), 111 deletions(-) delete mode 100644 benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml index d35086dcc..dcd6753fa 100644 --- a/.github/configs/amd-master.yaml +++ b/.github/configs/amd-master.yaml @@ -553,30 +553,6 @@ minimaxm2.5-fp8-mi355x-atom: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } -minimaxm2.5-fp4-mi355x-atom: - image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post - model: amd/MiniMax-M2.5-MXFP4 - model-prefix: minimaxm2.5 - runner: mi355x - precision: fp4 - framework: atom - multinode: false - seq-len-configs: - - isl: 1024 - osl: 1024 - search-space: - - { tp: 1, conc-start: 4, conc-end: 1024 } - - { tp: 2, conc-start: 4, conc-end: 1024 } - - { tp: 4, conc-start: 4, conc-end: 128 } - - { tp: 8, conc-start: 4, conc-end: 16 } - - isl: 8192 - osl: 1024 - search-space: - - { tp: 1, conc-start: 4, conc-end: 1024 } - - { tp: 2, conc-start: 4, conc-end: 1024 } - - { tp: 4, conc-start: 4, conc-end: 128 } - - { tp: 8, conc-start: 4, conc-end: 16 } - minimaxm2.5-fp4-mi355x-vllm: image: vllm/vllm-openai-rocm:v0.19.1 model: amd/MiniMax-M2.5-MXFP4 diff --git a/benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh b/benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh deleted file mode 100644 index ca84f8228..000000000 --- a/benchmarks/single_node/minimaxm2.5_fp4_mi355x_atom.sh +++ /dev/null @@ -1,80 +0,0 @@ -#!/usr/bin/env bash - -source "$(dirname "$0")/../benchmark_lib.sh" - -check_env_vars \ - MODEL \ - TP \ - CONC \ - ISL \ - OSL \ - RANDOM_RANGE_RATIO \ - RESULT_FILENAME \ - EP_SIZE \ - DP_ATTENTION - -if [[ -n "$SLURM_JOB_ID" ]]; then - echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME" -fi - -echo "TP: $TP, CONC: $CONC, ISL: $ISL, OSL: $OSL, EP_SIZE: $EP_SIZE, DP_ATTENTION: $DP_ATTENTION" - -SERVER_LOG=/workspace/server.log -PORT=${PORT:-8888} - -export OMP_NUM_THREADS=1 - -# Calculate max-model-len based on ISL and OSL -if [ "$ISL" = "1024" ] && [ "$OSL" = "1024" ]; then - CALCULATED_MAX_MODEL_LEN="" -else - CALCULATED_MAX_MODEL_LEN=" --max-model-len 10240 " -fi - -if [ "$EP_SIZE" -gt 1 ]; then - EP=" --enable-expert-parallel" -else - EP=" " -fi - -# Start GPU monitoring (power, temperature, clocks every second) -start_gpu_monitor - -set -x - -python3 -m atom.entrypoints.openai_server \ - --model $MODEL \ - --server-port $PORT \ - -tp $TP \ - --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \ - --trust-remote-code \ - > $SERVER_LOG 2>&1 & - -SERVER_PID=$! - -# Wait for server to be ready -wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID" - -export PYTHONDONTWRITEBYTECODE=1 -run_benchmark_serving \ - --model "$MODEL" \ - --port "$PORT" \ - --backend vllm \ - --input-len "$ISL" \ - --output-len "$OSL" \ - --random-range-ratio "$RANDOM_RANGE_RATIO" \ - --num-prompts "$((CONC * 10))" \ - --max-concurrency "$CONC" \ - --result-filename "$RESULT_FILENAME" \ - --result-dir /workspace/ \ - --trust-remote-code - -# After throughput, run evaluation only if RUN_EVAL is true -if [ "${RUN_EVAL}" = "true" ]; then - run_eval --framework lm-eval --port "$PORT" - append_lm_eval_summary -fi - -# Stop GPU monitoring -stop_gpu_monitor -set +x 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 61a7d0b61..9d6035a8b 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1813,13 +1813,6 @@ - "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-fp4-mi355x-atom - description: - - "Add MiniMax-M2.5 MXFP4 MI355X Atom benchmark (rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post)" - - "Single-node sweep: TP1–TP8, 1k/1k and 8k/1k ISL/OSL" - pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1042 - - config-keys: - minimaxm2.5-fp8-mi355x-atom description: From 6d4278bd6661256c2a55bc772809d7053e2c5b0e Mon Sep 17 00:00:00 2001 From: seungrokj Date: Mon, 27 Apr 2026 21:15:45 +0900 Subject: [PATCH 8/8] Fill in perf-changelog description for minimaxm2.5-fp8-mi355x-atom Co-Authored-By: Claude Sonnet 4.6 --- perf-changelog.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 9d6035a8b..f3fbf6bd9 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1818,4 +1818,4 @@ 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 \ No newline at end of file + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1194