TorchTitan: One-stop PyTorch native solution for production ready LLM pre-training

TL;DR

TorchTitan is a comprehensive, open-source PyTorch system that simplifies large language model training by unifying advanced techniques, enabling efficient scaling, and providing tools for production readiness and recipe comparison.

Contribution

It introduces TorchTitan, a modular, scalable, and interoperable system that integrates state-of-the-art LLM training techniques within PyTorch, streamlining development and optimization.

Findings

Achieved up to 65.08% acceleration with 1D parallelism at 128-GPU scale.

Demonstrated additional speedups of 12.59% and 30% with 2D and 3D parallelism at larger scales.

Validated performance improvements on Llama 3.1 models from 8B to 405B parameters.

Abstract

The development of large language models (LLMs) has been instrumental in advancing state-of-the-art natural language processing applications. Training LLMs with billions of parameters and trillions of tokens require sophisticated distributed systems that enable composing and comparing several state-of-the-art techniques in order to efficiently scale across thousands of accelerators. However, existing solutions are complex, scattered across multiple libraries/repositories, lack interoperability, and are cumbersome to maintain. Thus, curating and empirically comparing training recipes require non-trivial engineering effort. This paper introduces TorchTitan, an open-source, PyTorch-native distributed training system that unifies state-of-the-art techniques, streamlining integration and reducing overhead. TorchTitan enables 3D parallelism in a modular manner with elastic scaling, providing comprehensive logging, checkpointing, and debugging tools for production-ready training. It also incorporates hardware-software co-designed solutions, leveraging features like Float8 training and SymmetricMemory. As a flexible test bed, TorchTitan facilitates custom recipe curation and comparison, allowing us to develop optimized training recipes for Llama 3.1 and provide guidance on selecting techniques for maximum efficiency based on our experiences. We thoroughly assess TorchTitan on the Llama 3.1 family of LLMs, spanning 8 billion to 405 billion parameters, and showcase its exceptional performance, modular composability, and elastic scalability. By stacking training optimizations, we demonstrate accelerations of 65.08% with 1D parallelism at the 128-GPU scale (Llama 3.1 8B), an additional 12.59% with 2D parallelism at the 256-GPU scale (Llama 3.1 70B), and an additional 30% with 3D parallelism at the 512-GPU scale (Llama 3.1 405B) on NVIDIA H100 GPUs over optimized baselines.