FBI-LLM: Scaling Up Fully Binarized LLMs from Scratch via Autoregressive Distillation

TL;DR

This paper introduces FBI-LLM, a fully binarized large language model trained from scratch that matches the performance of full-precision models, using autoregressive distillation and revealing new training insights.

Contribution

It demonstrates the first successful training of large-scale fully binarized LLMs from scratch, matching full-precision performance without pretrained weights.

Findings

Binarized LLMs can be trained from scratch without pretrained weights.

FBI-LLM achieves competitive perplexity and task performance.

The training trajectory analysis provides new insights into binarized model training.

Abstract

This work presents a Fully BInarized Large Language Model (FBI-LLM), demonstrating for the first time how to train a large-scale binary language model from scratch (not the partial binary or ternary LLM like BitNet b1.58) to match the performance of its full-precision counterparts (e.g., FP16 or BF16) in transformer-based LLMs. It achieves this by employing an autoregressive distillation (AD) loss with maintaining equivalent model dimensions (130M, 1.3B, 7B) and training data volume as regular LLM pretraining, while delivering competitive results in terms of perplexity and task-specific effectiveness. Intriguingly, by analyzing the training trajectory, we find that the pretrained weight is not necessary for training binarized LLMs from scratch. This research encourages a new computational framework and may facilitate the future design of specialized hardware tailored for fully 1-bit LLMs. We make all models, code, and training dataset fully accessible and transparent to support further research (Code: https://github.com/LiqunMa/FBI-LLM. Model: https://huggingface.co/LiqunMa/).