[AMD] feat: MiniMax M3 day-zero benchmark for MI325X

.github/configs/amd-master.yaml

@@ -2847,6 +2847,35 @@ minimaxm3-fp8-mi355x-vllm-mtp:
       - { tp: 4, conc-start: 1, conc-end: 64, spec-decoding: mtp }
       - { tp: 8, ep: 8, dp-attn: true, conc-start: 128, conc-end: 256, spec-decoding: mtp }
 
+# MiniMax-M3 MXFP8 MI325X day-zero recipe. Reuse the dedicated ROCm image
+# and serving flags validated on MI355X, with the H200 search space: TP4 and
+# TP8 latency, TP4/TP8 expert parallelism, and TP8 data-parallel attention.
+minimaxm3-fp8-mi325x-vllm:
+  image: vllm/vllm-openai-rocm:minimax-m3
+  model: MiniMaxAI/MiniMax-M3-MXFP8
+  model-prefix: minimaxm3
+  runner: mi325x
+  precision: fp8
+  framework: vllm
+  multinode: false
+  scenarios:
+    fixed-seq-len:
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - { tp: 4, conc-start: 4, conc-end: 64 }
+      - { tp: 4, ep: 4, conc-start: 128, conc-end: 256 }
+      - { tp: 8, conc-start: 4, conc-end: 128 }
+      - { tp: 8, ep: 8, conc-start: 256, conc-end: 512 }
+      - { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 1024 }
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - { tp: 4, conc-start: 4, conc-end: 32 }
+      - { tp: 8, conc-start: 4, conc-end: 128 }
+      - { tp: 8, ep: 8, conc-start: 256, conc-end: 256 }
+      - { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 512 }
+
 # MiniMax-M3 MXFP8 MI300X day-zero recipe. Reuse the dedicated ROCm image and
 # MI355X serving shape, but retain the default BF16 KV cache because this
 # checkpoint lacks calibrated ROCm FP8 attention scales. Use the TP8-only H100

benchmarks/single_node/fixed_seq_len/minimaxm3_fp8_mi325x.sh

@@ -0,0 +1,85 @@
+#!/usr/bin/env bash
+
+# MiniMax-M3 MXFP8 MI325X (gfx942) single-node vLLM recipe.
+# https://recipes.vllm.ai/MiniMaxAI/MiniMax-M3?hardware=mi325x&variant=mxfp8
+# MXFP8 runs from TP=4 on gfx942; block size 128 is mandatory for MSA.
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    EP_SIZE \
+    DP_ATTENTION \
+    CONC \
+    ISL \
+    OSL \
+    MAX_MODEL_LEN \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+if [[ "$MODEL" != /* ]]; then hf download "$MODEL"; fi
+
+if [ -n "$ROCR_VISIBLE_DEVICES" ]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
+SERVER_LOG=/workspace/server.log
+export VLLM_ENGINE_READY_TIMEOUT_S=3600
+
+if [ "${EVAL_ONLY}" = "true" ]; then
+    setup_eval_context
+fi
+
+PARALLEL_ARGS=(--tensor-parallel-size "$TP")
+if [ "${DP_ATTENTION}" = "true" ]; then
+    PARALLEL_ARGS=(
+        --tensor-parallel-size 1
+        --data-parallel-size "$TP"
+        --enable-expert-parallel
+    )
+elif [ "$EP_SIZE" -gt 1 ]; then
+    PARALLEL_ARGS+=(--enable-expert-parallel)
+fi
+
+start_gpu_monitor
+
+set -x
+vllm serve "$MODEL" --port "$PORT" \
+    "${PARALLEL_ARGS[@]}" \
+    --block-size 128 \
+    --language-model-only \
+    --max-model-len "$MAX_MODEL_LEN" \
+    --attention-backend TRITON_ATTN \
+    --no-enable-prefix-caching \
+    --tool-call-parser minimax_m3 \
+    --reasoning-parser minimax_m3 \
+    --enable-auto-tool-choice > "$SERVER_LOG" 2>&1 &
+
+SERVER_PID=$!
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+run_benchmark_serving \
+    --model "$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 "$RESULT_FILENAME" \
+    --result-dir /workspace/ \
+    --trust-remote-code
+
+if [ "${RUN_EVAL}" = "true" ]; then
+    run_eval --framework lm-eval --port "$PORT"
+    append_lm_eval_summary
+fi
+
+stop_gpu_monitor
+set +x

perf-changelog.yaml

@@ -3745,3 +3745,12 @@
     - "H100-aligned layouts and concurrency ranges: TP8 and TP8+EP8 across 1k1k and 8k1k"
     - "Fix launch_mi300x-amds.sh node exclusion to use the current short Slurm node name"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1746
+
+- config-keys:
+    - minimaxm3-fp8-mi325x-vllm
+  description:
+    - "Initial submission: MiniMax-M3 MXFP8 day-zero single-node vLLM benchmark on MI325X / gfx942"
+    - "Follows the official MI325X MXFP8 recipe with vllm/vllm-openai-rocm:minimax-m3, block size 128, TRITON_ATTN, and MiniMax-M3 tool/reasoning parsers; fixed-sequence text benchmarking adds language-model-only, the default BF16 KV cache, disabled prefix caching, and a scenario-specific max model length"
+    - "H200-aligned layouts and concurrency ranges: TP4 and TP8 latency, TP4/TP8 expert parallelism, and TP8 data-parallel attention across 1k1k and 8k1k"
+    - "Route the MI325X Hugging Face cache and runtime compiler caches to node-local storage, and mount ROCm GPU devices explicitly"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1748

runners/launch_mi325x-amds.sh

@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
 set -euo pipefail
 
-export HF_HUB_CACHE_MOUNT="/nfsdata/sa/gharunner/gharunners/hf-hub-cache/"
+export HF_HUB_CACHE_MOUNT="/local-nvme/hf-hub-cache/"
 
 PARTITION="compute"
 SQUASH_FILE="/nfsdata/sa/gharunner/gharunners/squash/$(echo "$IMAGE" | sed 's/[\/:@#]/_/g').sqsh"
@@ -17,6 +17,8 @@ if [ -z "$JOB_ID" ]; then
 fi
 
 export PORT=$(( 40000 + (JOB_ID % 10000) ))
+export XDG_CACHE_HOME="/tmp/xdg-cache-$JOB_ID"
+export TRITON_CACHE_DIR="/tmp/triton-cache-$JOB_ID"
 
 trap 'rc=$?; scancel "$JOB_ID" 2>/dev/null || true; exit "$rc"' EXIT
 
@@ -34,7 +36,7 @@ srun --jobid="$JOB_ID" --job-name="$RUNNER_NAME" bash -c "
 "
 srun --jobid="$JOB_ID" \
 --container-image="$SQUASH_FILE" \
---container-mounts="$GITHUB_WORKSPACE:/workspace/,$HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE" \
+--container-mounts="$GITHUB_WORKSPACE:/workspace/,$HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE,/dev/kfd:/dev/kfd,/dev/dri:/dev/dri" \
 --container-mount-home \
 --container-writable \
 --container-remap-root \