inference-cache-config/qwen-moe.json

{
   "Qwen/Qwen3-30B-A3B-Instruct-2507": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "num_cores": 16,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "num_cores": 32,
      "auto_cast_type": "bf16"
    }
   ]
}