ProtGPT3-1.3B is a single-sequence autoregressive protein language model for protein sequence generation. It is part of the ProtGPT3 family, an open-source suite of promptable and aligned protein language models ranging from 112M to 10B parameters. ProtGPT3 models use a causal Mixtral-style Mixture-of-Experts architecture and are trained for causal language modeling on protein sequences.

The single-sequence ProtGPT3 models can generate proteins in either N-to-C or C-to-N direction using special directional tokens. The model is intended for unconditional or prefix-conditioned protein sequence generation and can be used as a base model for downstream protein design workflows.

Developed by: Anonymous authors
Model type: Autoregressive protein language model; causal decoder-only Mixture-of-Experts model
Language(s): Protein sequences / amino-acid sequences
License: More Information Needed
Finetuned from model: Not applicable / pretrained from scratch

Model Sources

Repository: https://huggingface.co/protgpt3
Paper: ProtGPT3: an Open-source family of Promptable and Aligned Protein Language Models
Code: https://anonymous.4open.science/r/protGPT3-2053/README.md

Uses

Direct Use

ProtGPT3-1.3B can be used for autoregressive generation of protein sequences. Users can generate sequences unconditionally or condition generation on an amino-acid prefix.

Downstream Use

The model may be fine-tuned or incorporated into protein design workflows, including family-specific generation, protein variant generation, and computational screening pipelines.

Out-of-Scope Use

The model should not be used as the sole basis for experimental, clinical, environmental, or safety-critical decisions. Generated proteins require downstream computational and experimental validation. The model is not guaranteed to generate functional, soluble, safe, or synthesizable proteins.

Bias, Risks, and Limitations

ProtGPT3-1.3B learns from public protein sequence datasets and may reproduce biases present in those datasets. Generated sequences may be low-complexity, nonfunctional, unstable, insoluble, or biologically implausible. Protein generation models may also present dual-use risks if used irresponsibly.

Recommendations

Users should apply appropriate computational filters, expert review, and experimental validation before using generated sequences. Users should also consider responsible-use practices for generative protein design.

How to Get Started with the Model

Install dependencies:

pip install transformers accelerate torch

Load the model and tokenizer:

python

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "protgpt3/ProtGPT3-1.3B"  # Replace with the final checkpoint name

# Load tokenizer for generation
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True,add_bos_token=True, add_eos_token=False, padding_side="left")

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

model.eval()

Generate a protein sequence

python

import torch

prompt = ""  # Optionally provide an amino-acid prefix or model-specific direction

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    output_ids = model.generate(
        inputs["input_ids"],
        max_new_tokens=512,
        do_sample=True,
        temperature=0.8,
        top_p=0.9,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.pad_token_id,
    )

sequence = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print(sequence) # output includes directional token "1" or "2" to denote if sequence was generated N-to-C or C-to-N

Generate from an amino-acid prefix

python

import torch

# forward N-to-C generation with special token "1" 
prefix = "1MKT" # use special token "2" instead of "1" for reverse  C-to-N generation

inputs = tokenizer(prefix, return_tensors="pt").to(model.device)

with torch.no_grad():
    output_ids = model.generate(
        inputs["input_ids"],
        max_new_tokens=256,
        do_sample=True,
        temperature=0.8,
        top_p=0.9,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.eos_token_id,
    )

sequence = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print(sequence)

Batch generation

python

import torch

prompts = [
    "",
    "1MKT", # N-to-C generation
    "2MAV", # C-to-N generation
]

inputs = tokenizer(
    prompts,
    return_tensors="pt",
    padding=True,
).to(model.device)

with torch.no_grad():
    output_ids = model.generate(
        inputs["input_ids"],
        max_new_tokens=256,
        do_sample=True,
        temperature=0.8,
        top_p=0.9,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.bos_token_id,
    )

sequences = tokenizer.batch_decode(output_ids, skip_special_tokens=True)

for sequence in sequences:
    print(sequence)

Training Details

Training Data

ProtGPT3-1.3B was trained on publicly available protein sequence data from UniRef90 and the GigaRef subset of the Dayhoff Atlas. The 1.3B-parameter model used approximately 15M UniRef90 sequences and 28M GigaRef sequences, corresponding to approximately 9.8B training tokens.

Training Procedure

Preprocessing

Protein sequences were sampled from UniRef90 and GigaRef. During training, each sequence was assigned a generation direction, either N-to-C or C-to-N, with a special token prepended to indicate the direction.

Training Hyperparameters

Training regime: bfloat16
Architecture: Mixtral-style sparse Mixture-of-Experts causal decoder
Maximum sequence length: 1024
Optimizer: AdamW
Learning rate: 5e-4
Weight decay: 0.1
Gradient clipping: 1.0
Batch size: 500
Number of training GPUs: 4

Evaluation

Testing Data, Factors & Metrics

Testing Data

The model was evaluated on held-out protein sequences with at most 50% sequence identity to the training set. It was also benchmarked on ProteinGym.

Metrics

Evaluation included validation perplexity, sequence diversity, predicted pLDDT, proportion of terminating sequences, proportion of low-complexity sequences, and ProteinGym Spearman correlation.

Results

Larger ProtGPT3 single-sequence models showed improved perplexity, sequence quality, and diversity. ProtGPT3-1.3B serves as the smallest single-sequence model in the family and provides a computationally accessible checkpoint for protein generation.

Technical Specifications

Model Architecture and Objective

ProtGPT3-1.3B is a decoder-only causal language model using a Mixtral-style sparse Mixture-of-Experts architecture. It was trained with a causal language modeling objective on protein sequences.

Compute Infrastructure

Hardware

NVIDIA H100 GPUs.

Software

Training used FlashAttention-2, online mini-batch packing, Liger Kernel, and DeepSpeed.

Citation

BibTeX:

bibtex

@article{protgpt3,
  title={ProtGPT3: an Open-source family of Promptable and Aligned Protein Language Models},
  author={Anonymous Authors},
  year={2026}
}

More Information

All models and code are released through the Hugging Face ecosystem and accompanying code repository.

Model Card Authors

Anonymous authors

Model Card Contact

Anonymous authors