inference-cache-config/llama.json

{
   "meta-llama/Meta-Llama-3.1-8B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "num_cores": 2,
      "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": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 16,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 32,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    }
  ],
  "meta-llama/Llama-2-7b-hf": [
    {
      "batch_size": 1,
      "sequence_length": 2048,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    }
  ],
  "meta-llama/Llama-2-13b-hf": [
    {
      "batch_size": 1,
      "sequence_length": 2048,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    }
  ],
   "meta-llama/Meta-Llama-3-8B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    }
  ],
   "meta-llama/Llama-3.2-1B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    }
  ],
   "meta-llama/Llama-3.2-3B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "bf16"
    }
  ]
}