jbenbudd/adpr-llama

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:May 28, 2026Architecture:Transformer Warm

The jbenbudd/adpr-llama is a 7 billion parameter LLaMA-based model, specifically LoRA-fine-tuned from GreatCaptainNemo/ProLLaMA_Stage_1. It is designed for the specialized task of predicting ADP-ribosylation (ADPr) post-translational modification sites. This model excels at identifying ADPr sites within 21-residue peptide windows, outputting results in a structured format like 'Sites='.

Loading preview...

Overview

The jbenbudd/adpr-llama is a specialized language model, a 7 billion parameter variant of the LLaMA architecture. It has been fine-tuned using the LoRA (Low-Rank Adaptation) method, building upon the GreatCaptainNemo/ProLLaMA_Stage_1 base model.

Key Capabilities

  • ADP-ribosylation (ADPr) Site Prediction: The primary function of this model is to accurately predict ADPr post-translational modification sites.
  • Peptide Window Analysis: It processes 21-residue peptide windows to identify potential ADPr sites.
  • Structured Output: Predictions are formatted clearly, indicating the specific sites found (e.g., Sites=<R5,D12,...>).

Training and Evaluation

This model card represents a training-only stub. Comprehensive evaluation metrics, including ROC, accuracy, precision, recall, F1 score, and confusion matrix, are intended to be generated and filled in by a companion evaluation notebook after the model is run on a held-out test set.

Good For

  • Researchers and developers working in proteomics or molecular biology who need to identify ADPr sites.
  • Integrating into bioinformatics pipelines for automated ADPr prediction from peptide sequences.
  • Specific applications requiring highly specialized biological sequence analysis, particularly for post-translational modifications.