• Added support for converting a PyTorch binary model to a TensorFlow checkpoint
• Added support for converting CKPT to safetensors

python -m venv .

# linux
source bin/activate

# windows
.\Scripts\activate

## Install

pip install -r requirements.txt

python .\bin2ckpt.py --model_path="PATH TO MODEL REPO" --checkpoint_path="PATH TO OUTPUT FILE"

I was lazy so you have to edit the path in the script file lmao

python ckpt2safetensors.py

详细说明可以参见《将 PyTorch 版 bin 模型转换成 Tensorflow 版 ckpt》

bin_path = './pretrained_model/pytorch_model/'
bin_model = 'pytorch_model.bin'
ckpt_path = './pretrained_model/tensorflow_model/'
ckpt_model = 'bert_model.ckpt'

convert(bin_path, bin_model, ckpt_path, ckpt_model)

• bin_path: pytorch model path
• bin_model: pytorch model name
• ckpt_path: path to save tf ckpt
• ckpt_model: tf ckpt name

Notice: this script only supports to convert the BERT model. If you need to convert other models, please modify the function to_tf_var_name() and variable tensors_to_transpose.

• SpanBERT

  • SpanBERT (base & cased): 12-layer, 768-hidden, 12-heads , 110M parameters
  • SpanBERT (large & cased): 24-layer, 1024-hidden, 16-heads, 340M parameters

  Download: GoogleDrive | BaiduDrive (Code: wtyr) | CTDrive

• SimCSE

  • supervised SimCSE (base & uncased)
  • supervised SimCSE (large & uncased)

  Download: GoogleDrive | BaiduDrive (Code: gnq3) | CTDrive