onca-2.0-12b-gguf

<p align="center"> <img src="./assets/onca-logo-horizontal.svg" alt="Onca 2.0 logo" width="520"> </p>

ONCA 2.0 12B GGUF

<p align="center"> <img src="./assets/onca20-holdout-primary-comparison.png" alt="ONCA 2.0 four-task primary-holdout comparison" width="900"> </p> <p align="center">_{BF16 reference scores on the unchanged 1,309-example primary holdout; compare values within each task panel.}</p>

Summary

ONCA 2.0 is an open oncology language model for trial screening, clinical reasoning, pathology extraction, and variant evidence interpretation. It builds on google/gemma-4-12B-it with continued supervised fine-tuning on a provenance-labeled oncology corpus while retaining the four-task ONCA 1.5 evaluation contract.

This page provides BF16, Q8_0, Q6_K, and Q4_K_M GGUF files generated from the merged ONCA 2.0 checkpoint. All variants are collected here so users can choose the precision and memory tier suited to their hardware.

At a Glance

Quick Start

Use a recent llama.cpp build with Gemma 4 support. The Q4_K_M file is a practical starting point for lower-memory systems.

llama-cli \
  -m onca-2.0-12b-int4-gguf.gguf \
  -cnv \
  -p 'Patient: metastatic pancreatic adenocarcinoma; ECOG 1. Trial: metastatic PDAC, ECOG 0-1. Return JSON with keys eligible, reason, and missing_information.'

Replace the model path with the BF16, Q8_0, or Q6_K file when more memory and higher fidelity are available. Checksums and validation details are recorded in verification_20260716.json.

Training Scope

The source BF16 checkpoint was trained on 25,302 examples. Validation, test, and primary-holdout sets retain the ONCA 1.5 four-task evaluation contract.

Related Releases

• onca-2.0-12B: BF16 Transformers reference release.
• onca-2.0-12B-INT8: 8-bit BitsAndBytes release.
• onca-2.0-12B-INT4: 4-bit BitsAndBytes release.
• onca-2.0-12B-GGUF: combined GGUF collection (this page).

Limitations

• This is a research model, not a clinical decision system.
• Outputs require review by qualified experts before real-world use.
• Structured or parser-valid output does not guarantee factual correctness.
• The benchmark figure reports the BF16 reference checkpoint; individual GGUF quantizations were not fully rebenchmarked.
• Lower-bit quantization can change output quality and behavior.

Citation

A formal ONCA 2.0 citation will be added with the accompanying manuscript. Until then, cite this model repository, the exact GGUF file, and its quantization.

Acknowledgements

ONCA 2.0 continues the ONCA project lineage and builds on Google Gemma, llama.cpp, and the open-data contributors whose datasets supported training and evaluation.