inference-cache-config/granite.json

{
   "ibm-granite/granite-3.1-2b-instruct": [
    {
      "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": 1,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 32,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    }
  ],
  "ibm-granite/granite-3.1-8b-instruct": [
   {
      "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": 1,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    },
    {
      "batch_size": 32,
      "sequence_length": 4096,
      "num_cores": 8,
      "auto_cast_type": "bf16"
    }
  ]
}