inference-cache-config/trn1/llama3.json

{
   "meta-llama/Meta-Llama-3.1-8B": [
    {
      "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": 4,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 16,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 32,
      "sequence_length": 4096,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    }
  ],
  "meta-llama/Llama-2-7b-hf": [
    {
      "batch_size": 1,
      "sequence_length": 2048,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    }
  ],
  "meta-llama/Llama-2-13b-hf": [
    {
      "batch_size": 1,
      "sequence_length": 2048,
      "tensor_parallel_size": 8,
      "instance_type" : "trn1"
    }
  ],
   "meta-llama/Meta-Llama-3-8B": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    }
  ],
   "meta-llama/Llama-3.2-1B": [
    {
      "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"
    }
  ],
   "meta-llama/Llama-3.2-3B": [
    {
      "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"
    }
  ],
     "TinyLlama/TinyLlama-1.1B-Chat-v1.0": [
    {
      "batch_size": 1,
      "sequence_length": 2048,
      "tensor_parallel_size": 2,
      "instance_type" : "trn1"
    }
  ],
  "meta-llama/Llama-3.3-70B-Instruct": [
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 24,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "tensor_parallel_size": 24,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 32,
      "instance_type" : "trn1"
    },
    {
      "batch_size": 8,
      "sequence_length": 4096,
      "tensor_parallel_size": 32,
      "instance_type" : "trn1"
    }
  ]
}