{
  "hfId": "MiniMaxAI/MiniMax-M2.7",
  "modelRevision": "main",
  "hardware": {
    "hwClass": "DISCRETE_GPU",
    "gpuName": "Intel Arc Pro B70",
    "gpuCount": 4,
    "vramGb": 32,
    "cpu": "AMD EPYC 9015",
    "ramGb": 16,
    "os": "Ubuntu 24.04.4"
  },
  "engineName": "vllm",
  "engineVersion": "0.20.1-local",
  "quantization": "INT4 AutoRound W4A16",
  "promptTokens": 512,
  "outputTokens": 128,
  "contextLength": 2048,
  "batchSize": 1,
  "tokSOut": 20.11,
  "tokSTotal": 100.53815829779805,
  "peakVramGb": 28.11,
  "notes": "Lasimeri/MiniMax-M2.7-int4-AutoRound target weights submitted under base MiniMaxAI/MiniMax-M2.7 because quant repo lookup previously failed. 4x Intel Arc Pro B70 TP4, local vLLM INC/XPU FusedMoE WNA16 patch, CCL_ZE_IPC_EXCHANGE=pidfd, hybrid B70 MoE config: tuned M=1 int4_w4a16 plus default prompt-size keys 64/256/512. No power-limit changes.",
  "engineFlags": {
    "commandSnippet": "VLLM_TUNED_CONFIG_FOLDER=/home/steve/bench-results/minimax-m2.7-autoround-vllm/moe-config-hybrid-m1-default-prefill CCL_ZE_IPC_EXCHANGE=pidfd CCL_TOPO_P2P_ACCESS=1 vllm bench throughput --backend vllm --model /mnt/corsair-external/llm-models/minimax-m2.7-int4-autoround --tokenizer /mnt/corsair-external/llm-models/minimax-m2.7-int4-autoround --trust-remote-code --dtype bfloat16 --tensor-parallel-size 4 --distributed-executor-backend mp --max-model-len 2048 --max-num-batched-tokens 1024 --max-num-seqs 1 --dataset-name random --random-input-len 512 --random-output-len 128 --random-range-ratio 0 --num-prompts 1 --disable-log-stats",
    "tensorParallel": 4,
    "gpuLayers": 0,
    "flashAttn": true,
    "contextLength": 2048,
    "attentionBackend": "flash_attn",
    "concurrency": 1,
    "extraFlags": "XPU/Level Zero, vLLM local INC FusedMoE WNA16 patch, VLLM_TUNED_CONFIG_FOLDER hybrid B70 MoE config, CCL_ATL_TRANSPORT=ofi, CCL_ZE_IPC_EXCHANGE=pidfd, CCL_TOPO_P2P_ACCESS=1, VLLM_XPU_ENABLE_XPU_GRAPH=0",
    "engineQuant": "inc",
    "gpuMemUtil": 0.9
  }
}