CosyVoice/runtime/triton_trtllm/model_repo/token2wav_dit/config.pbtxt

# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

name: "token2wav_dit"
backend: "python"
max_batch_size: ${triton_max_batch_size}

dynamic_batching {
    max_queue_delay_microseconds: ${max_queue_delay_microseconds}
    priority_levels: 10
    default_priority_level: 10
}

parameters: { key: "FORCE_CPU_ONLY_INPUT_TENSORS" value: {string_value:"no"}}
parameters [
  {
   key: "model_dir",
   value: {string_value:"${model_dir}"}
  }
]

input [
  {
    name: "target_speech_tokens"
    data_type: TYPE_INT32
    dims: [-1]
  },
  {
    name: "reference_wav"
    data_type: TYPE_FP32
    dims: [-1]
  },
  {
    name: "reference_wav_len"
    data_type: TYPE_INT32
    dims: [1]
  },
  {
    name: "finalize"
    data_type: TYPE_BOOL
    dims: [ 1 ]
    reshape: { shape: [ ] }
    optional: true
  },
  {
    name: "conformer_cnn_cache"
    data_type: TYPE_FP16
    dims: [ 512, -1 ]
    optional: true
  },
  {
    name: "conformer_att_cache"
    data_type: TYPE_FP16
    dims: [ 10, 8, -1, 128 ]
    optional: true
  },
  {
    name: "estimator_cnn_cache"
    data_type: TYPE_FP16
    dims: [ 10, 16, -1, 1024, 2 ]
    optional: true
  },
  {
    name: "estimator_att_cache"
    data_type: TYPE_FP16
    dims: [ 10, 16, -1, 8, -1, 128 ]
    optional: true
  },
  {
    name: "mel"
    data_type: TYPE_FP32
    dims: [ 80, -1 ]
    optional: true
  },
  {
    name: "source"
    data_type: TYPE_FP32
    dims: [ 1, -1 ]
    optional: true
  },
  {
    name: "speech"
    data_type: TYPE_FP32
    dims: [ -1 ]
    optional: true
  }
]
output [
  {
    name: "waveform"
    data_type: TYPE_FP32
    dims: [ -1 ]
  },
  {
    name: "conformer_cnn_cache"
    data_type: TYPE_FP16
    dims: [ 512, -1 ]
  },
  {
    name: "conformer_att_cache"
    data_type: TYPE_FP16
    dims: [ 10, 8, -1, 128 ]
  },
  {
    name: "estimator_cnn_cache"
    data_type: TYPE_FP16
    dims: [ 10, 16, -1, 1024, 2 ]
  },
  {
    name: "estimator_att_cache"
    data_type: TYPE_FP16
    dims: [ 10, 16, -1, 8, -1, 128 ]
  },
  {
    name: "mel"
    data_type: TYPE_FP32
    dims: [ 80, -1 ]
  },
  {
    name: "source"
    data_type: TYPE_FP32
    dims: [ 1, -1 ]
  },
  {
    name: "speech"
    data_type: TYPE_FP32
    dims: [ -1 ]
  }
]

instance_group [
  {
    count: 1
    kind: KIND_CPU
  }
]