|
--- |
|
license: mit |
|
language: |
|
- en |
|
pipeline_tag: summarization |
|
library_name: transformers.js |
|
--- |
|
https://huggingface.co/JustinDu/BARTxiv with ONNX weights to be compatible with Transformers.js. |
|
|
|
## Steps to get ONNX quantized weights |
|
1. Use optimum-cli to convert the PyTorch weights to ONNX format. |
|
|
|
```bash |
|
optimum-cli export onnx --model JustinDu/BARTxiv BARTxiv_onnx |
|
``` |
|
|
|
2. Use the `quantize` script in the Transformers.js repo to quantize the ONNX weights. |
|
|
|
```bash |
|
python -m scripts.quantize --input_folder BARTxiv_onnx --output_folder BARTxiv_onnx/onnx |
|
``` |
|
|
|
Move the `.onnx` files in the `BARTxiv_onnx` folder to `BARTxiv_onnx/onnx` (these are the FP32 weights). |
