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

# Triton should expect as input a single string
# input of variable length named 'text_input'
input [
  {
    name: "text_input"
    data_type: TYPE_STRING
    dims: [ -1 ]
  }
]


# output of variable length named 'embedding_outputt'
output [
  {
    name: "embedding_output"
    data_type: TYPE_FP32 # 또는 TYPE_FP16
    dims: [ -1, -1 ]     # [seq_len, hidden_size]
  }
]

parameters: [
  {
    key: "base_model_path",
    value: {string_value: "/cheetah/input/model/groupuser/codesage-large-v2"}
  }
]

instance_group [
  {
    kind: KIND_AUTO
    count: 1
  }
]