BiLLM icon indicating copy to clipboard operation
BiLLM copied to clipboard

Tool for converting LLMs from uni-directional to bi-directional by removing causal mask for tasks like classification and sentence embeddings. Compatible with 🤗 transformers.

Results 3 BiLLM issues
Sort by recently updated
recently updated
newest added

- fix bug: attention_mask=None - upgrade transformers

Hi, Thank you for really nice piece of work. How can i do sequence classification with this? since you only provide example for NER.

feature

Hi, Thanks for sharing this very interesting work. I had a question about how the bidirectional attention mask is implemented [here](https://github.com/WhereIsAI/BiLLM/blob/main/src/billm/modeling_mistral.py#L141) Based on this implementation, it seems like even the...