inference-cache-config/qwen2.5-large.json

{
  "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    }
  ],
  "Qwen/Qwen2.5-32B-Instruct": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    }
  ],
  "Qwen/Qwen2.5-72B-Instruct": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "num_cores": 24,
      "auto_cast_type": "bf16"
    }
  ]
}