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