Página inicial Explorar Ajuda
Entrar
github-repo
/
CosyVoice
mirror de https://ghfast.top/https://github.com/FunAudioLLM/CosyVoice
2
0
Fork 1
Arquivos Issues 0 Wiki
Tree: 2a3e033ee1
Branches Tags
CosyVoice
/
cosyvoice
/
bin
/
export_trt.sh

export_trt.sh 3.1 KB
Histórico Raw

12345678910111213141516171819202122232425 
  1. #!/bin/bash
    2. 

  2. # Copyright 2024 Alibaba Inc. All Rights Reserved.
    3. 

  3. # download tensorrt from https://developer.nvidia.com/tensorrt/download/10x, check your system and cuda for compatibability
    4. 

  4. # for example for linux + cuda12.4, you can download https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.0.1/tars/TensorRT-10.0.1.6.Linux.x86_64-gnu.cuda-12.4.tar.gz
    5. 

  5. TRT_DIR=<YOUR_TRT_DIR>
    6. 

  6. MODEL_DIR=<YOUR_MODEL_DIR>
    7. 

  7. export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$TRT_DIR/lib:/usr/local/cuda/lib64
    8. 

  8. 

  9. # cosyvoice export
    10. 

  10. $TRT_DIR/bin/trtexec --onnx=$MODEL_DIR/flow.decoder.estimator.fp32.onnx --saveEngine=$MODEL_DIR/flow.decoder.estimator.fp32.mygpu.plan --minShapes=x:2x80x4,mask:2x1x4,mu:2x80x4,cond:2x80x4 --optShapes=x:2x80x200,mask:2x1x200,mu:2x80x200,cond:2x80x200 --maxShapes=x:2x80x3000,mask:2x1x3000,mu:2x80x3000,cond:2x80x3000 --inputIOFormats=fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw --outputIOFormats=fp32:chw
    11. 

  11. $TRT_DIR/bin/trtexec --onnx=$MODEL_DIR/flow.decoder.estimator.fp32.onnx --saveEngine=$MODEL_DIR/flow.decoder.estimator.fp16.mygpu.plan --fp16 --minShapes=x:2x80x4,mask:2x1x4,mu:2x80x4,cond:2x80x4 --optShapes=x:2x80x200,mask:2x1x200,mu:2x80x200,cond:2x80x200 --maxShapes=x:2x80x3000,mask:2x1x3000,mu:2x80x3000,cond:2x80x3000 --inputIOFormats=fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw --outputIOFormats=fp16:chw
    12. 

  12. 

  13. # cosyvoice2 export with cache
    14. 

  14. $TRT_DIR/bin/trtexec --onnx=$MODEL_DIR/flow.decoder.estimator.fp32.onnx --saveEngine=$MODEL_DIR/flow.decoder.estimator.fp32.mygpu.plan \
    15. 

  15.     --minShapes=x:2x80x4,mask:2x1x4,mu:2x80x4,cond:2x80x4,down_blocks_kv_cache:1x4x2x0x512x2,mid_blocks_kv_cache:12x4x2x0x512x2,up_blocks_kv_cache:1x4x2x0x512x2 \
    16. 

  16.     --optShapes=x:2x80x200,mask:2x1x200,mu:2x80x200,cond:2x80x200,down_blocks_kv_cache:1x4x2x100x512x2,mid_blocks_kv_cache:12x4x2x100x512x2,up_blocks_kv_cache:1x4x2x100x512x2 \
    17. 

  17.     --maxShapes=x:2x80x1500,mask:2x1x1500,mu:2x80x1500,cond:2x80x1500,down_blocks_kv_cache:1x4x2x200x512x2,mid_blocks_kv_cache:12x4x2x200x512x2,up_blocks_kv_cache:1x4x2x200x512x2 \
    18. 

  18.     --inputIOFormats=fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw \
    19. 

  19.     --outputIOFormats=fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw,fp32:chw
    20. 

  20. $TRT_DIR/bin/trtexec --onnx=$MODEL_DIR/flow.decoder.estimator.fp32.onnx --saveEngine=$MODEL_DIR/flow.decoder.estimator.fp16.mygpu.plan --fp16 \
    21. 

  21.     --minShapes=x:2x80x4,mask:2x1x4,mu:2x80x4,cond:2x80x4,down_blocks_kv_cache:1x4x2x0x512x2,mid_blocks_kv_cache:12x4x2x0x512x2,up_blocks_kv_cache:1x4x2x0x512x2 \
    22. 

  22.     --optShapes=x:2x80x200,mask:2x1x200,mu:2x80x200,cond:2x80x200,down_blocks_kv_cache:1x4x2x100x512x2,mid_blocks_kv_cache:12x4x2x100x512x2,up_blocks_kv_cache:1x4x2x100x512x2 \
    23. 

  23.     --maxShapes=x:2x80x1500,mask:2x1x1500,mu:2x80x1500,cond:2x80x1500,down_blocks_kv_cache:1x4x2x200x512x2,mid_blocks_kv_cache:12x4x2x200x512x2,up_blocks_kv_cache:1x4x2x200x512x2 \
    24. 

  24.     --inputIOFormats=fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw \
    25. 

  25.     --outputIOFormats=fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw,fp16:chw
    26.