# BinGE: TODO

TODO: 2 line summary  and link to paper

## Usage
```python
import torch
from transformers import AutoTokenizer, AutoModel, AutoConfig
from peft import PeftModel


if __name__ == "__main__":
    # Loading base Meta-Llama-3 model, along with custom code that enables bidirectional connections in decoder-only LLMs.
    tokenizer = AutoTokenizer.from_pretrained(
        "McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp"
    )
    config = AutoConfig.from_pretrained(
        "McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp", trust_remote_code=True
    )
    model = AutoModel.from_pretrained(
        "McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp",
        trust_remote_code=True,
        config=config,
        torch_dtype=torch.bfloat16,
        device_map="cuda" if torch.cuda.is_available() else "cpu",
    )

    # Loading MNTP (Masked Next Token Prediction) model.
    model = PeftModel.from_pretrained(
        model,
        "McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp",
    )

    model = model.merge_and_unload()  # This can take several minutes on cpu

    # Loading BinGSE model. This loads the trained LoRA weights on top of MNTP model. Hence the final weights are -- Base model + MNTP (LoRA) + BinGSE (LoRA).
    model = PeftModel.from_pretrained(
        model, model_path 
    )

```

TODO: initialize wrapper, provide example to check loading happened properly - see https://huggingface.co/McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-unsup-simcse