92 lines
1.2 KiB
Plaintext
92 lines
1.2 KiB
Plaintext
# Triton backend to use
|
|
name: "Llama-3.2-1B-Instruct"
|
|
backend: "python"
|
|
max_batch_size: 4
|
|
|
|
# Triton should expect as input a single string
|
|
# input of variable length named 'text_input'
|
|
input [
|
|
|
|
{
|
|
name: "input0"
|
|
data_type: TYPE_FP32
|
|
dims: [ 1, 1, 1, 1 ]
|
|
|
|
|
|
reshape {
|
|
shape: [ 11, 11, 11 ]
|
|
}
|
|
|
|
},
|
|
{
|
|
name: "input1"
|
|
data_type: TYPE_UINT64
|
|
dims: [ 2, 2, 2, 2 ]
|
|
|
|
|
|
reshape {
|
|
shape: [ 22, 22, 22, 22 ]
|
|
}
|
|
|
|
}
|
|
|
|
]
|
|
|
|
|
|
# Triton should expect to respond with a single string
|
|
# output of variable length named 'text_output'
|
|
output [
|
|
|
|
{
|
|
name: "output0"
|
|
data_type: TYPE_BOOL
|
|
dims: [ 1 ]
|
|
|
|
},
|
|
{
|
|
name: "output1"
|
|
data_type: TYPE_FP64
|
|
dims: [ 1, 100 ]
|
|
|
|
reshape {
|
|
shape: [ 100 ]
|
|
}
|
|
|
|
},
|
|
{
|
|
name: "text_output"
|
|
data_type: TYPE_STRING
|
|
dims: [ -1 ]
|
|
|
|
}
|
|
|
|
]
|
|
|
|
parameters: [
|
|
{
|
|
key: "base_model_path",
|
|
value: {string_value: "/cheetah/input/model/groupuser/Llama-3.2-1B-Instruct"}
|
|
},
|
|
{
|
|
key: "is_adapter_model",
|
|
value: {string_value: "false"}
|
|
},
|
|
{
|
|
key: "adapter_model_path",
|
|
value: {string_value: ""}
|
|
},
|
|
|
|
{
|
|
key: "quantization",
|
|
value: {string_value: "int4"}
|
|
}
|
|
]
|
|
|
|
instance_group [
|
|
{
|
|
kind: KIND_GPU
|
|
count: 1
|
|
}
|
|
]
|
|
|