docs

游雁

2023-04-21 14a14d0e40e48942fe03241532b81636062a3420

docs

 docs/academic_recipe/asr_recipe.md

 docs/academic_recipe/lm_recipe.md

 docs/academic_recipe/punc_recipe.md

 docs/academic_recipe/sd_recipe.md

 docs/academic_recipe/sv_recipe.md

 docs/academic_recipe/vad_recipe.md

 docs/index.rst

@@ -22,13 +22,13 @@

.. toctree::
   :maxdepth: 1
   :caption: Academic Recipe
   :caption: Academic Egs

   ./recipe/asr_recipe.md
   ./recipe/punc_recipe.md
   ./recipe/vad_recipe.md
   ./recipe/sv_recipe.md
   ./recipe/sd_recipe.md
   ./academic_recipe/asr_recipe.md
   ./academic_recipe/punc_recipe.md
   ./academic_recipe/vad_recipe.md
   ./academic_recipe/sv_recipe.md
   ./academic_recipe/sd_recipe.md


.. toctree::
@@ -40,7 +40,7 @@

.. toctree::
   :maxdepth: 1
   :caption: ModelScope pipeline
   :caption: ModelScope Egs

   ./modescope_pipeline/quick_start.md
   ./modescope_pipeline/asr_pipeline.md
@@ -52,13 +52,13 @@

.. toctree::
   :maxdepth: 1
   :caption: Huggingface pipeline
   :caption: Huggingface Egs

   Undo

.. toctree::
   :maxdepth: 1
   :caption: funasr library
   :caption: Funasr Library

   ./build_task.md


 funasr/export/README.md

@@ -16,7 +16,7 @@
pip install onnx onnxruntime # Optional, for onnx quantization
```

## Export model
## Usage
   `Tips`: torch>=1.11.0

   ```shell
@@ -37,7 +37,46 @@

   `fallback-num`: specify the number of fallback layers to perform automatic mixed precision quantization.

## Performance Benchmark of Runtime

### Export onnx format model
#### Export model from modelscope
```shell
python -m funasr.export.export_model --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type onnx
```
#### Export model from local path
The model'name must be `model.pb`
```shell
python -m funasr.export.export_model --model-name /mnt/workspace/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type onnx
```
#### Test onnx model
Ref to [test](https://github.com/alibaba-damo-academy/FunASR/tree/main/funasr/export/test)

### Export torchscripts format model
#### Export model from modelscope
```shell
python -m funasr.export.export_model --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type torch
```

#### Export model from local path
The model'name must be `model.pb`
```shell
python -m funasr.export.export_model --model-name /mnt/workspace/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type torch
```
#### Test onnx model
Ref to [test](https://github.com/alibaba-damo-academy/FunASR/tree/main/funasr/export/test)

## Runtime
### ONNXRuntime
#### ONNXRuntime-python
Ref to [docs](https://alibaba-damo-academy.github.io/FunASR/en/runtime/onnxruntime_python.html)
#### ONNXRuntime-cpp
Ref to [docs](https://alibaba-damo-academy.github.io/FunASR/en/runtime/onnxruntime_cpp.html)
### Libtorch
#### Libtorch-python
Ref to [docs](https://alibaba-damo-academy.github.io/FunASR/en/runtime/libtorch_python.html)
#### Libtorch-cpp
Undo
## Performance Benchmark

### Paraformer on CPU

@@ -48,27 +87,6 @@
### Paraformer on GPU
[nv-triton](https://github.com/alibaba-damo-academy/FunASR/tree/main/funasr/runtime/triton_gpu)

## For example
### Export onnx format model
Export model from modelscope
```shell
python -m funasr.export.export_model --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type onnx
```
Export model from local path, the model'name must be `model.pb`.
```shell
python -m funasr.export.export_model --model-name /mnt/workspace/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type onnx
```

### Export torchscripts format model
Export model from modelscope
```shell
python -m funasr.export.export_model --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type torch
```

Export model from local path, the model'name must be `model.pb`.
```shell
python -m funasr.export.export_model --model-name /mnt/workspace/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type torch
```

## Acknowledge
Torch model quantization is supported by [BladeDISC](https://github.com/alibaba/BladeDISC), an end-to-end DynamIc Shape Compiler project for machine learning workloads. BladeDISC provides general, transparent, and ease of use performance optimization for TensorFlow/PyTorch workloads on GPGPU and CPU backends. If you are interested, please contact us.