liushiliushi/ConfTuner-LLaMA

Llama 7B Uncertainty Calibration Model (Brier Loss)

This model is a fine-tuned version of Llama-3.1-8B-Instruct optimized for uncertainty calibration using our method ConfTuner: Training Large Language Models to Express Their Confidence Verbally, accepted by NeurIPS 2025.

Model Details

Model Description

• Developed by: liushiliushi
• Model type: Llama fine-tuned with PEFT/LoRA
• Language(s): English
• License: Same as base model (Llama 3.1)
• Finetuned from model: meta-llama/Llama-3.1-8B-Instruct

Uses

Direct Use

This model is optimized for tasks requiring well-calibrated uncertainty estimates.

Training Details

Training Hyperparameters

• Learning rate: 1e-5
• Epochs: 2
• Loss function: Brier score
• Batch size: 16

Framework versions

• PEFT 0.12.0