HALO: Hadamard-Assisted Lower-Precision Optimization for LLMs
This addresses the problem of efficient and accurate low-precision fine-tuning for LLMs, which is incremental as it builds on existing quantization methods but introduces novel techniques to overcome specific bottlenecks.
The paper tackles the challenge of quantized training for Large Language Models (LLMs) by introducing HALO, a quantization-aware training approach that achieves near-full-precision-equivalent results during fine-tuning on various tasks, with up to 1.41x end-to-end speedup on RTX 4090 GPUs.
Quantized training of Large Language Models (LLMs) remains an open challenge, as maintaining accuracy while performing all matrix multiplications in low precision has proven difficult. This is particularly the case when fine-tuning pre-trained models, which can have large weight and activation outlier values that make lower-precision optimization difficult. To address this, we present HALO, a novel quantization-aware training approach for Transformers that enables accurate and efficient low-precision training by combining 1) strategic placement of Hadamard rotations in both forward and backward passes, which mitigate outliers, 2) high-performance kernel support, and 3) FSDP integration for low-precision communication. Our approach ensures that all large matrix multiplications during the forward and backward passes are executed in lower precision. Applied to LLAMA-family models, HALO achieves near-full-precision-equivalent results during fine-tuning on various tasks, while delivering up to 1.41x end-to-end speedup for full fine-tuning on RTX 4090 GPUs. HALO efficiently supports both standard and parameterefficient fine-tuning (PEFT). Our results demonstrate the first practical approach to fully quantized LLM fine-tuning that maintains accuracy in 8-bit precision, while delivering performance benefits. Code is available at https://github.com/IST-DASLab/HALO.