LitCab: Lightweight Language Model Calibration over Short- and Long-form Responses

TL;DR

LitCab is a lightweight, parameter-efficient calibration method for language models that significantly improves their probability estimates across diverse text generation tasks, enhancing trustworthiness without extensive fine-tuning.

Contribution

The paper introduces LitCab, a simple linear layer-based calibration method that outperforms existing techniques in calibrating large language models with minimal additional parameters.

Findings

Larger models are better calibrated on short tasks but not on longer ones.

GPT models are better calibrated than LLaMA and Vicuna, despite fewer parameters.

Limited fine-tuning can worsen calibration, emphasizing the importance of training setup.

Abstract

A model is considered well-calibrated when its probability estimate aligns with the actual likelihood of the output being correct. Calibrating language models (LMs) is crucial, as it plays a vital role in detecting and mitigating hallucinations of LMs as well as building more trustworthy models. However, standard calibration techniques may not be suited for LM calibration. For instance, post-processing methods such as temperature scaling do not reorder the candidate generations. On the other hand, training-based methods require fine-tuning the entire model, which is impractical for LMs of large scale. We present LitCab, a lightweight calibration mechanism consisting of a single linear layer that takes the input text representation and predicts a bias term, which is then added to the LM output logits. LitCab improves model calibration by only adding < 2% of the original model parameters. For evaluation, we construct CaT, a benchmark consisting of eight text generation tasks, covering responses ranging from short phrases to paragraphs. We test LitCab with Llama2-7B, where it improves calibration across all tasks, reducing the average ECE score by as large as 30%. We further conduct a comprehensive evaluation with multiple popular open-sourced LMs from GPT and LLaMA families, yielding the following key findings: (i) Larger models within the same family exhibit better calibration on tasks with short generation tasks, but not necessarily for longer ones. (ii) GPT-family models show superior calibration compared to LLaMA, Llama2, and Vicuna models, despite having much fewer parameters. (iii) Fine-tuning pretrained model (e.g., LLaMA) with samples of limited purpose (e.g., conversations) may lead to worse calibration, highlighting the importance of fine-tuning setups for calibrating LMs.