Model Card for peleke-phi-4

This model is a fine-tuned version of microsoft/phi-4 for antibody sequence generation. It takes in an antigen sequence, and returns novel Fv portions of heavy and light chain antibody sequences.

Quick start

1. Load in the Model

model_name = 'silicobio/peleke-phi-4'
config = PeftConfig.from_pretrained(model_name)

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, torch_dtype=torch.bfloat16, trust_remote_code=True).cuda()
model.resize_token_embeddings(len(tokenizer))
model = PeftModel.from_pretrained(model, model_name).cuda()

2. Format your Input

This model uses <epi> and </epi> to annotate epitope residues of interest.

It may be easier to use other characters for annotation, such as [ ]'s. For example: ...CSFS[S][F][V]L[N]WY.... Then, use the following function to properly format the input.

def format_prompt(antigen_sequence):
    epitope_seq = re.sub(r'\[([A-Z])\]', r'<epi>\1</epi>', antigen_sequence)
    formatted_str = f"Antigen: {epitope_seq}<|im_end|>\nAntibody:"
    return formatted_str

3. Generate an Antibody Sequence

prompt = format_prompt(antigen)
inputs = tokenizer(prompt, return_tensors="pt")
inputs = {k: v.cuda() for k, v in inputs.items()}

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=1000,
        do_sample=True,
        temperature=0.7,
        pad_token_id=tokenizer.eos_token_id,
        use_cache=False,
    )

full_text = tokenizer.decode(outputs[0], skip_special_tokens=False)
antibody_sequence = full_text.split('<|im_end|>')[1].replace('Antibody: ', '')
print(f"Antigen: {antigen}\nAntibody: {antibody_sequence}\n")

This will generate a |-delimited output, which is an Fv portion of a heavy and light chain.

Antigen: NPPTFSPALL...
Antibody: QVQLVQSGGG...|DIQMTQSPSS...

Training procedure

This model was trained with SFT.

Framework versions

• PEFT 0.17.0
• TRL: 0.19.1
• Transformers: 4.54.0
• Pytorch: 2.7.1
• Datasets: 4.0.0
• Tokenizers: 0.21.2