# Triton Backend for Python.
backend: "python"
max_batch_size: 0

input [
  {
    name: "text_input"
    data_type: TYPE_STRING
    dims: [ 1 ]
  },
  {
    name: "max_length"
    data_type: TYPE_INT32
    dims: [ 1 ]
  },
  {
    name: "max_new_tokens"
    data_type: TYPE_INT32
    dims: [ 1 ]
  },
  {
    name: "do_sample"
    data_type: TYPE_BOOL
    dims: [ 1 ]
  },
  {
    name: "top_k"
    data_type: TYPE_INT32
    dims: [ 1 ]
  },
  {
    name: "top_p"
    data_type: TYPE_FP32
    dims: [ 1 ]
  },
  {
    name: "temperature"
    data_type: TYPE_FP32
    dims: [ 1 ]
  },
  {
    name: "repetition_penalty"
    data_type: TYPE_FP32
    dims: [ 1 ]
  },
  {
    name: "stream"
    data_type: TYPE_BOOL
    dims: [ 1 ]
  }
]

output [
  {
    name: "text_output"
    data_type: TYPE_STRING
    dims: [ 1 ]
  }
]


parameters: [
  {
    key: "model_path",
    value: {string_value: "/cheetah/input/model/groupuser/TinyLlama-1.1B-Chat-v1.0"}
  }
]


instance_group [
  {
    kind: KIND_AUTO
    count: 1
  }
]