[AMD] [AgentX] Add MiniMax-M3 MXFP8 EAGLE3 on MI325X

benchmarks/single_node/agentic/minimaxm3_fp8_mi325x_mtp.sh

@@ -0,0 +1,107 @@
+#!/usr/bin/env bash
+set -eo pipefail
+set -x
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+export EVAL_FRAMEWORK="lm-eval"
+
+check_env_vars MODEL TP CONC KV_OFFLOADING RESULT_DIR DURATION EP_SIZE DP_ATTENTION PORT EVAL_ONLY
+
+DRAFT_MODEL="Inferact/MiniMax-M3-EAGLE3-GQA"
+NUM_SPEC_TOKENS=3
+SYNTHETIC_ACCEPT_LEN=2.78
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+    echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+if [[ -n "$ROCR_VISIBLE_DEVICES" ]]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
+if [[ -n "$MODEL_PATH" ]]; then
+    if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$MODEL" --local-dir "$MODEL_PATH"
+    fi
+else
+    hf download "$MODEL"
+    export MODEL_PATH="$MODEL"
+fi
+hf download "$DRAFT_MODEL"
+
+rocm-smi || true
+amd-smi || true
+
+resolve_trace_source
+install_agentic_deps
+
+SERVER_LOG="$RESULT_DIR/server.log"
+mkdir -p "$RESULT_DIR"
+
+SERVER_PID=""
+cleanup_agentic_services() {
+    local exit_code=$?
+    trap - EXIT INT TERM
+    set +e
+    stop_background_process_tree "$SERVER_PID" "vLLM server" 60
+    exit "$exit_code"
+}
+trap cleanup_agentic_services EXIT
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
+require_agentic_kv_offload_none
+export AIPERF_SERVER_METRICS_URLS="http://localhost:${PORT}/metrics"
+export AIPERF_REQUIRED_SERVER_METRIC_PREFIX="vllm:"
+
+if [ "$EVAL_ONLY" = "true" ]; then
+    SPEC_CONFIG="{\"method\": \"eagle3\", \"model\": \"$DRAFT_MODEL\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"attention_backend\": \"TRITON_ATTN\"}"
+else
+    SPEC_CONFIG="{\"method\": \"eagle3\", \"model\": \"$DRAFT_MODEL\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"attention_backend\": \"TRITON_ATTN\", \"rejection_sample_method\": \"synthetic\", \"synthetic_acceptance_length\": $SYNTHETIC_ACCEPT_LEN}"
+fi
+
+export PYTHONNOUSERSITE=1
+export VLLM_ENGINE_READY_TIMEOUT_S=3600
+export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=1800
+export VLLM_USE_BREAKABLE_CUDAGRAPH=0
+
+VLLM_CMD=(
+    vllm serve "$MODEL_PATH"
+    --served-model-name "$MODEL"
+    --host 0.0.0.0
+    --port "$PORT"
+    --tensor-parallel-size "$TP"
+    --gpu-memory-utilization 0.90
+    --kv-cache-dtype fp8
+    --block-size 128
+    --language-model-only
+    --attention-backend TRITON_ATTN
+    --enable-prefix-caching
+    --enable-chunked-prefill
+    --max-num-batched-tokens 32768
+    --max-num-seqs "$((2 * CONC))"
+    --speculative-config "$SPEC_CONFIG"
+    --tool-call-parser minimax_m3
+    --reasoning-parser minimax_m3
+    --enable-auto-tool-choice
+    --default-chat-template-kwargs '{"thinking_mode":"enabled"}'
+    --trust-remote-code
+    --stream-interval 20
+)
+write_command "$RESULT_DIR/server_command.txt" "${VLLM_CMD[@]}"
+"${VLLM_CMD[@]}" > "$SERVER_LOG" 2>&1 &
+SERVER_PID=$!
+
+wait_for_ready \
+    --endpoint "http://0.0.0.0:${PORT}/health" \
+    --log "$SERVER_LOG" \
+    --pid "$SERVER_PID"
+
+if [ "$EVAL_ONLY" = "true" ]; then
+    run_eval --port "$PORT"
+else
+    build_replay_cmd "$RESULT_DIR"
+    REPLAY_CMD+=" --apply-chat-template"
+    run_agentic_replay_and_write_outputs "$RESULT_DIR"
+fi

configs/amd-master.yaml

@@ -1565,6 +1565,20 @@ minimaxm3-fp8-mi325x-vllm-agentic:
       - { tp: 8, ep: 8, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, conc-list: [10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 32] }
       - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "0.3.11.post1" }, conc-list: [24, 32, 36, 40, 44, 48, 52, 56, 60, 64, 72, 80, 96], router: { name: vllm-router, version: "0.1.14" } }
 
+minimaxm3-fp8-mi325x-vllm-agentic-mtp:
+  image: vllm/vllm-openai-rocm:v0.27.1
+  model: MiniMaxAI/MiniMax-M3-MXFP8
+  model-prefix: minimaxm3
+  runner: cluster:mi325x-amds
+  precision: fp8
+  framework: vllm
+  multinode: false
+  scenarios:
+    agentic-coding:
+    - dram-utilization: 0.20
+      search-space:
+      - { tp: 8, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 2, 4, 8, 10, 12, 14, 16, 18] }
+
 minimaxm3-fp4-mi355x-vllm-agentic:
   image: vllm/vllm-openai-rocm:nightly-dcfebf93f4eccf30f71872283331eee757915daf
   model: amd/MiniMax-M3-MXFP4

perf-changelog.yaml

@@ -1738,7 +1738,7 @@
     - "TP=2 and TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths"
     - "Add --gpu-memory-utilization 0.9 to server launch"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1133
-  
+
 
 - config-keys:
     - dsv4-fp8-h200-vllm
@@ -1860,7 +1860,6 @@
     - "Image pinned to lmsysorg/sglang:deepseek-v4-b300@sha256:26e116bd211e300dbb76924d56c5cbe6cc3ee5ee2fe314859cb8774f5bc070f3"
     - "DP-attention path enables SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN=1 for better SWA eviction behavior"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1185
-
 
 - config-keys:
     - dsv4-fp4-b200-sglang
@@ -5910,6 +5909,14 @@
     - "Replace the TP8/EP8 low-concurrency arm (conc [1,2,4]) hicache offload with kv-offloading: none to reduce per-request latency at low load; conc [1,2,4] are now tested on both tp=4+hicache and tp=8+no-offload so SemiAnalysis can select the Pareto-optimal point per concurrency"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2570
 
+- config-keys:
+    - minimaxm3-fp8-mi325x-vllm-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Add MiniMax-M3 MXFP8 AgentX on MI325X with vLLM v0.27.1, EAGLE3-GQA, golden synthetic acceptance length 2.78, and the measured resident TP8 frontier through its c16-c18 latency knee."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2579
+
 - config-keys:
     - qwen3.5-fp4-b200-sglang-mtp
   scenario-type:
@@ -5928,7 +5935,7 @@
     - "Use native EAGLE MTP (3 steps, top-k 1, 4 draft tokens) and golden synthetic acceptance length 2.49 for throughput; eval retains real verification."
     - "Follow the official SGLang DeepSeek-V4 Blackwell recipe, require nonempty SGLang server metrics, and keep pooled AgentX connections alive across inter-turn gaps."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2578
-  
+
 - config-keys:
     - dsv4-fp4-b300-sglang-agentic-hicache-mtp
   scenario-type:
@@ -5938,4 +5945,3 @@
     - "Use native EAGLE MTP (3 steps, top-k 1, 4 draft tokens) and golden synthetic acceptance length 2.49 for throughput; eval retains real verification."
     - "Follow the official SGLang DeepSeek-V4 Blackwell recipe, require nonempty SGLang server metrics, keep pooled AgentX connections alive, let AIPerf own HiCache warmup, and reserve transient MoE workspace at DEP8 c512."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2577
-