inference-cache-config/llama-variants.json

{
  "princeton-nlp/Sheared-LLaMA-1.3B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "fp16"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "fp16"
    }
  ],
  "lmsys/vicuna-7b-v1.5": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "fp16"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "num_cores": 2,
      "auto_cast_type": "fp16"
    }
  ]
}