[codex] add MiniMax M3 FP4 MI355X vLLM MTP benchmark

.github/configs/amd-master.yaml

@@ -2667,6 +2667,36 @@ minimaxm3-fp4-mi355x-vllm:
       - { tp: 4, conc-start: 1, conc-end: 128 }
       - { tp: 8, ep: 8, dp-attn: true, conc-start: 128, conc-end: 512 }
 
+# EAGLE3 speculative-decoding variant of minimaxm3-fp4-mi355x-vllm. Pair the
+# amd/MiniMax-M3-MXFP4 target with Inferact/MiniMax-M3-EAGLE3 and three draft
+# tokens. Search space mirrors the MI355X MXFP8 MTP entry, trimming the base
+# FP4 sweep at extreme concurrency where speculative decoding loses value.
+minimaxm3-fp4-mi355x-vllm-mtp:
+  image: vllm/vllm-openai-rocm:nightly-3f5a1e1733200760169ff31ebe60a271072b199e
+  model: amd/MiniMax-M3-MXFP4
+  model-prefix: minimaxm3
+  runner: mi355x
+  precision: fp4
+  framework: vllm
+  multinode: false
+  scenarios:
+    fixed-seq-len:
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - { tp: 8, conc-start: 1, conc-end: 64, spec-decoding: mtp }
+      - { tp: 8, ep: 8, conc-start: 1, conc-end: 256, spec-decoding: mtp }
+      - { tp: 4, conc-start: 1, conc-end: 64, spec-decoding: mtp }
+      - { tp: 4, ep: 4, conc-start: 64, conc-end: 256, spec-decoding: mtp }
+      - { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 512, spec-decoding: mtp }
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - { tp: 8, conc-start: 1, conc-end: 64, spec-decoding: mtp }
+      - { tp: 8, ep: 8, conc-start: 1, conc-end: 256, spec-decoding: 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 MXFP4 MI355X atom recipe:
 # https://github.com/ROCm/ATOM/blob/5d42d49f9e4292e5b61475917e92e7ec1b1dacb7/recipes/MiniMax-M3.md
 # block size 128 is mandatory for MSA. TP4 on a single gfx950 node, per the recipe.

benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_mi355x_vllm_mtp.sh

@@ -0,0 +1,98 @@
+#!/usr/bin/env bash
+
+# MiniMax-M3 MXFP4 MI355X (gfx950) single-node vLLM recipe with EAGLE3
+# speculative decoding. This is the spec-decoding=mtp variant of
+# minimaxm3_fp4_mi355x_vllm.sh and uses three speculative tokens from
+# Inferact/MiniMax-M3-EAGLE3. The pinned nightly includes upstream AMD
+# MiniMax-M3 SupportsEagle3 support, so no runtime model patch is needed.
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    EP_SIZE \
+    DP_ATTENTION \
+    CONC \
+    ISL \
+    OSL \
+    MAX_MODEL_LEN \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME
+
+DRAFT_MODEL="Inferact/MiniMax-M3-EAGLE3"
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+if [[ "$MODEL" != /* ]]; then hf download "$MODEL"; fi
+hf download "$DRAFT_MODEL"
+
+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
+export VLLM_USE_BREAKABLE_CUDAGRAPH=0
+
+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 [ "$EP_SIZE" -gt 1 ]; then
+    PARALLEL_ARGS+=(--enable-expert-parallel)
+fi
+
+NUM_SPEC_TOKENS=3
+
+start_gpu_monitor
+
+set -x
+vllm serve "$MODEL" --port "$PORT" \
+    "${PARALLEL_ARGS[@]}" \
+    --trust-remote-code \
+    --block-size 128 \
+    --no-enable-prefix-caching \
+    --language-model-only \
+    --max-model-len "$MAX_MODEL_LEN" \
+    --attention-backend TRITON_ATTN \
+    --speculative-config "{\"method\": \"eagle3\", \"model\": \"$DRAFT_MODEL\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS}" \
+    --tool-call-parser minimax_m3 \
+    --enable-auto-tool-choice \
+    --reasoning-parser minimax_m3 > "$SERVER_LOG" 2>&1 &
+
+SERVER_PID=$!
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+pip install -q datasets pandas
+
+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 \
+    --use-chat-template
+
+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

@@ -4213,3 +4213,11 @@
     - "Image: lmsysorg/sglang:nightly-dev-cu13-20260608-303757cc."
     - "8k/1k STP recipes: 1P1D TP4 (conc 1-256), 5P1D DEP4+1D DEP16 (conc 2048, NIXL), 6P1D and 7P1D DEP4+1D DEP16 (conc 5120, Mooncake)."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1921
+
+- config-keys:
+    - minimaxm3-fp4-mi355x-vllm-mtp
+  description:
+    - "Add a MiniMax-M3 MXFP4 MI355X vLLM benchmark with EAGLE3 speculative decoding using amd/MiniMax-M3-MXFP4 and Inferact/MiniMax-M3-EAGLE3 with three speculative tokens."
+    - "Reuse the pinned vllm/vllm-openai-rocm:nightly-3f5a1e1733200760169ff31ebe60a271072b199e image, text-only target path, TRITON_ATTN, automatic tool choice, MiniMax-M3 parsers, VLLM_USE_BREAKABLE_CUDAGRAPH=0, default KV-cache dtype, and automatic MoE backend selection."
+    - "Pass --use-chat-template for MTP acceptance and mirror the existing MiniMax-M3 MXFP8 MI355X MTP TP/EP/DP-attention search space at 1k1k and 8k1k."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1939