ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

doc fix (#4829)

Browse source
This commit is contained in:
Zhentao Chen 2022-06-15 08:23:01 +08:00 committed by GitHub
parent 04542bc6db
commit 7dcd79366f
1 changed files with 6 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
docs/readthedocs/source/doc/Nano/QuickStart/pytorch_inference.md
View file

@ -92,10 +92,10 @@ By default, `Trainer.quantize()` doesn't search the tuning space and returns the
### Quantization using Intel Neural Compressor ### Quantization using Intel Neural Compressor
By default, Intel Neural Compressor is not installed with BigDL-Nano. So if you determine to use it as your quantization backend, you'll need to install it first: By default, Intel Neural Compressor is not installed with BigDL-Nano. So if you determine to use it as your quantization backend, you'll need to install it first:
```shell ```shell
# We have tested on neural-compressor>=1.8.1,<=1.11.0 pip install neural-compressor==1.11.0
pip install 'neural-compressor>=1.8.1,<=1.11.0'
``` ```
**Quantization without extra accelerator** **Quantization without extra accelerator**
Without extra accelerator, `Trainer.quantize()` returns a pytorch module with desired precision and accuracy. Following the example in [Runtime Acceleration](#runtime-acceleration), you can add quantization as below: Without extra accelerator, `Trainer.quantize()` returns a pytorch module with desired precision and accuracy. Following the example in [Runtime Acceleration](#runtime-acceleration), you can add quantization as below:
```python ```python
q_model = trainer.quanize(model, calib_dataloader=dataloader) q_model = trainer.quanize(model, calib_dataloader=dataloader)
@ -109,8 +109,9 @@ trainer.predict(q_model, dataloader)
``` ```
This is a most basic usage to quantize a model with defaults, INT8 precision, and without search tuning space to control accuracy drop. This is a most basic usage to quantize a model with defaults, INT8 precision, and without search tuning space to control accuracy drop.
**Quantization with ONNXRuntime accelerator** **Quantization with ONNXRuntime accelerator**
Without the ONNXRuntime accelerator, `Trainer.quantize()` will return a model with compressed precision but running inference in the ONNXRuntime engine. If your INC version >= 1.11, it's also required to install onnxruntime-extensions as a dependency of INC when using ONNXRuntime as backend as well as the dependencies required in [ONNXRuntime Acceleration](#onnxruntime-acceleration):
Without the ONNXRuntime accelerator, `Trainer.quantize()` will return a model with compressed precision but running inference in the ONNXRuntime engine. It's also required to install onnxruntime-extensions as a dependency of INC when using ONNXRuntime as backend as well as the dependencies required in [ONNXRuntime Acceleration](#onnxruntime-acceleration):
```shell ```shell
pip install onnx onnxruntime onnxruntime-extensions pip install onnx onnxruntime onnxruntime-extensions
``` ```