---
base_model:
- leaBroe/HeavyBERTa
- leaBroe/LightGPT
pipeline_tag: translation
---
# Heavy2Light
Heavy2Light is an seq2seq model designed to generate light chain antibody sequences from corresponding heavy chain inputs. It leverages [HeavyBERTa](https://huggingface.co/leaBroe/HeavyBERTa) as the encoder and [LightGPT](https://huggingface.co/leaBroe/LightGPT) as the decoder. The model is fine-tuned on paired antibody chain data from the [OAS](https://opig.stats.ox.ac.uk/webapps/oas/) and [PLAbDab](https://opig.stats.ox.ac.uk/webapps/plabdab/) databases. The model utilizes [Adapters](https://github.com/adapter-hub/adapters) for efficient fine-tuning. You can either download the full model weights and adapter from this repository, or directly use the Heavy2Light adapter available in its dedicated [directory](https://huggingface.co/leaBroe/Heavy2Light_adapter) on Hugging Face.  
For more information, please visit our GitHub [repository](https://github.com/ibmm-unibe-ch/Heavy2Light.git).  

## How to use the model
```python
from transformers import EncoderDecoderModel, AutoTokenizer, GenerationConfig
from adapters import init

model_path = "leaBroe/Heavy2Light"
subfolder_path = "heavy2light_final_checkpoint"

model = EncoderDecoderModel.from_pretrained(model_path)

tokenizer = AutoTokenizer.from_pretrained(model_path, subfolder=subfolder_path)

init(model)
adapter_name = model.load_adapter("leaBroe/Heavy2Light_adapter", set_active=True)
model.set_active_adapters(adapter_name)

generation_config = GenerationConfig.from_pretrained(model_path)

# example input heavy sequence
heavy_seq = "QLQVQESGPGLVKPSETLSLTCTVSGASSSIKKYYWGWIRQSPGKGLEWIGSIYSSGSTQYNPALGSRVTLSVDTSQTQFSLRLTSVTAADTATYFCARQGADCTDGSCYLNDAFDVWGRGTVVTVSS"

inputs = tokenizer(
    heavy_seq,
    padding="max_length",
    truncation=True,
    max_length=250,
    return_tensors="pt"
)

generated_seq = model.generate(
    input_ids=inputs.input_ids,
    attention_mask=inputs.attention_mask,
    num_return_sequences=1,
    output_scores=True,
    return_dict_in_generate=True,
    generation_config=generation_config,
    bad_words_ids=[[4]],
    do_sample=True,
    temperature=1.0,
)

generated_text = tokenizer.decode(
    generated_seq.sequences[0],
    skip_special_tokens=True,
)

print("Generated light sequence:", generated_text)
```