Liger Kernel: Efficient Triton Kernels for LLM Training

TL;DR

Liger-Kernel introduces optimized Triton kernels for LLM training, significantly improving throughput and reducing memory usage, with a focus on modularity, compatibility, and accessibility.

Contribution

This work presents Liger-Kernel, a set of optimized Triton kernels specifically designed for efficient large language model training, with notable performance improvements.

Findings

20% increase in training throughput

60% reduction in GPU memory usage

Compatibility across diverse environments

Abstract

Training Large Language Models (LLMs) efficiently at scale presents a formidable challenge, driven by their ever-increasing computational demands and the need for enhanced performance. In this work, we introduce Liger-Kernel, an open-sourced set of Triton kernels developed specifically for LLM training. With kernel optimization techniques like kernel operation fusing and input chunking, our kernels achieve on average a 20% increase in training throughput and a 60% reduction in GPU memory usage for popular LLMs compared to HuggingFace implementations. In addition, Liger-Kernel is designed with modularity, accessibility, and adaptability in mind, catering to both casual and expert users. Comprehensive benchmarks and integration tests are built in to ensure compatibility, performance, correctness, and convergence across diverse computing environments and model architectures. The source code is available under a permissive license at: github.com/linkedin/Liger-Kernel.