inference-cache-config/trn1/granite.json

{
   "ibm-granite/granite-3.1-2b-instruct": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 32,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    }
  ],
  "ibm-granite/granite-3.1-8b-instruct": [
   {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 4,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 32,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    }
  ]
}