inference-cache-config/trn1/llama4.json

{
  "meta-llama/Llama-4-Scout-17B-16E-Instruct": [
    {
      "task": "text-generation",
      "instance_type": "trn1",
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 16,
      "auto_cast_type": "bf16"
    },
    {
      "task": "text-generation",
      "instance_type": "trn1",
      "batch_size": 4,
      "sequence_length": 4096,
      "tensor_parallel_size": 16,
      "auto_cast_type": "bf16"
    },
    {
      "task": "text-generation",
      "instance_type": "trn1",
      "batch_size": 1,
      "sequence_length": 4096,
      "tensor_parallel_size": 32,
      "auto_cast_type": "bf16"
    },
    {
      "task": "text-generation",
      "instance_type": "trn1",
      "batch_size": 4,
      "sequence_length": 4096,
      "tensor_parallel_size": 32,
      "auto_cast_type": "bf16"
    },
    {
      "task": "text-generation",
      "instance_type": "trn1",
      "batch_size": 8,
      "sequence_length": 4096,
      "tensor_parallel_size": 32,
      "auto_cast_type": "bf16"
    }
  ]
}