add B300 vLLM AgentX single-node MiniMax-M3 FP4 EAGLE3-GQA MTP / 新增 B300 vLLM AgentX 单节点 MiniMax-M3 FP4 EAGLE3-GQA MTP

benchmarks/single_node/agentic/minimaxm3_fp4_b300_mtp.sh

@@ -0,0 +1,122 @@
+#!/usr/bin/env bash
+set -euo pipefail
+set -x
+
+# MiniMax-M3 NVFP4 B300 AgentX with EAGLE3-GQA and synthetic acceptance.
+# DRAM KV offload uses vLLM's SimpleCPUOffloadConnector in lazy mode.
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+export EVAL_FRAMEWORK="lm-eval"
+
+check_env_vars MODEL TP CONC KV_OFFLOADING TOTAL_CPU_DRAM_GB RESULT_DIR DURATION
+
+DRAFT_MODEL="Inferact/MiniMax-M3-EAGLE3-GQA"
+NUM_SPEC_TOKENS=3
+# Golden AL for the GQA draft head: golden_al_distribution/minimaxm3_eagle3_gqa.yaml
+# minimax-m3.thinking_on[3]. The non-GQA curve (minimaxm3_eagle3.yaml) reads 2.83
+# at the same level -- that head is not what this script runs.
+SYNTHETIC_ACCEPT_LEN=2.78
+
+if [[ -n "${SLURM_JOB_ID:-}" ]]; then
+    echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
+fi
+
+if [[ -n "${MODEL_PATH:-}" ]]; then
+    if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$MODEL" --local-dir "$MODEL_PATH"
+    fi
+    DRAFT_MODEL_PATH="/data/models/${DRAFT_MODEL##*/}"
+    if [[ ! -d "$DRAFT_MODEL_PATH" || -z "$(ls -A "$DRAFT_MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$DRAFT_MODEL" --local-dir "$DRAFT_MODEL_PATH"
+    fi
+else
+    hf download "$MODEL"
+    export MODEL_PATH="$MODEL"
+    hf download "$DRAFT_MODEL"
+    DRAFT_MODEL_PATH="$DRAFT_MODEL"
+fi
+
+nvidia-smi
+resolve_trace_source
+install_agentic_deps
+
+OFFLOAD_ARGS=()
+if require_agentic_kv_offload_backend vllm-simple; then
+    CPU_OFFLOAD_BYTES=$((TOTAL_CPU_DRAM_GB * 1024 * 1024 * 1024))
+    export VLLM_USE_SIMPLE_KV_OFFLOAD=1
+    OFFLOAD_CONFIG=$(printf \
+        '{"kv_connector":"SimpleCPUOffloadConnector","kv_role":"kv_both","kv_connector_extra_config":{"cpu_bytes_to_use":%d,"lazy_offload":true}}' \
+        "$CPU_OFFLOAD_BYTES")
+    OFFLOAD_ARGS=(--kv-transfer-config "$OFFLOAD_CONFIG")
+fi
+
+export PYTHONNOUSERSITE=1
+export VLLM_ENGINE_READY_TIMEOUT_S=3600
+export VLLM_FLOAT32_MATMUL_PRECISION=high
+export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm
+
+SERVER_LOG="$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
+
+if [ "${EVAL_ONLY:-}" = "true" ]; then
+    SPEC_CONFIG=$(printf \
+        '{"method":"eagle3","model":"%s","num_speculative_tokens":%d,"attention_backend":"FLASH_ATTN"}' \
+        "$DRAFT_MODEL_PATH" "$NUM_SPEC_TOKENS")
+else
+    SPEC_CONFIG=$(printf \
+        '{"method":"eagle3","model":"%s","num_speculative_tokens":%d,"attention_backend":"FLASH_ATTN","rejection_sample_method":"synthetic","synthetic_acceptance_length":%.2f}' \
+        "$DRAFT_MODEL_PATH" "$NUM_SPEC_TOKENS" "$SYNTHETIC_ACCEPT_LEN")
+fi
+
+{ set +x; } 2>/dev/null
+VLLM_CMD=(
+    vllm serve "$MODEL_PATH"
+    --served-model-name "$MODEL"
+    --host 0.0.0.0
+    --port "$PORT"
+    --tensor-parallel-size "$TP"
+    --gpu-memory-utilization 0.9
+    --block-size 128
+    --language-model-only
+    --enable-prefix-caching
+    --no-enable-flashinfer-autotune
+    --reasoning-parser minimax_m3
+    --tool-call-parser minimax_m3
+    --enable-auto-tool-choice
+    --default-chat-template-kwargs '{"thinking_mode":"enabled"}'
+    --attention-config '{"backend":"FLASHINFER","use_trtllm_attention":true,"indexer_kv_dtype":"fp8"}'
+    --kv-cache-dtype fp8
+    --max-cudagraph-capture-size 512
+    --max-num-batched-tokens 16384
+    --stream-interval 20
+    --trust-remote-code
+    --speculative-config "$SPEC_CONFIG"
+    "${OFFLOAD_ARGS[@]}"
+)
+printf '%q ' "${VLLM_CMD[@]}" | tee "$RESULT_DIR/vllm_command.txt"
+printf '\n' | tee -a "$RESULT_DIR/vllm_command.txt"
+"${VLLM_CMD[@]}" > "$SERVER_LOG" 2>&1 &
+SERVER_PID=$!
+echo "Server PID: $SERVER_PID"
+set -x
+
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+if [ "${EVAL_ONLY}" = "true" ]; then
+    run_eval --port "$PORT"
+else
+    build_replay_cmd "$RESULT_DIR"
+    run_agentic_replay_and_write_outputs "$RESULT_DIR"
+fi

configs/nvidia-master.yaml

@@ -7412,6 +7412,24 @@ qwen3.5-fp4-b200-sglang-agentic-mtp:
       - { tp: 2, ep: 1, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 4, 8, 12, 14] }
       - { tp: 2, ep: 1, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: hicache }, conc-list: [16, 18, 20, 22, 24, 28, 32] }
 
+minimaxm3-fp4-b300-vllm-agentic-mtp:
+  image: vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7
+  model: nvidia/MiniMax-M3-NVFP4
+  model-prefix: minimaxm3
+  runner: cluster:b300-nv
+  precision: fp4
+  framework: vllm
+  multinode: false
+  scenarios:
+    agentic-coding:
+    # 0.683 maps the B300 TP4 SimpleCPU point to a 1,024 GB
+    # engine-level CPU KV budget. GPU-resident points receive a zero budget.
+    - dram-utilization: 0.683
+      search-space:
+      - { tp: 4, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 2, 5, 10, 15, 20] }
+      - { tp: 2, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 2, 5] }
+      - { tp: 4, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [30, 40, 50, 60, 65, 70, 75] }
+
 dsv4-fp4-b200-sglang-agentic-hicache:
   image: lmsysorg/sglang:nightly-dev-cu13-20260707-b4155233
   model: deepseek-ai/DeepSeek-V4-Pro

perf-changelog.yaml

@@ -5449,3 +5449,14 @@
     - "Enable prefill-only INT4 quick-reduce: set VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4 and VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB=2048 on the prefill workers via a new prefill_env channel (mirrors the existing decode_env path in server_vllm.sh)."
     - "Cap the 1P1D TP4 concurrency sweep at 256 (was 512); drop the 2P1D TP4 layout (128/256/512) as it is CI-flaky with negligible curve impact."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1943
+
+- config-keys:
+    - minimaxm3-fp4-b300-vllm-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Add five MiniMax-M3 NVFP4 B300 single-node aggregate vLLM AgentX points: TP8 concurrency 1; TP4 concurrency 1, 2, and 16; TP2 concurrency 2."
+    - "Use the Inferact/MiniMax-M3-EAGLE3-GQA draft model with three speculative tokens and synthetic rejection sampling; thinking-on synthetic acceptance length is 2.78."
+    - "Use vllm/vllm-openai:nightly-387189c42997b27e2c04b5d97ef8190ffa2bf909 with prefix caching enabled, default KV-cache dtype, FlashInfer TRT-LLM attention with FP8 indexer KV, and the EAGLE3 drafter on FLASH_ATTN."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2328
+