Open SourceArtificial Intelligence PlatformLightning AI announced a partnership with Nvidia to release a newThunderThe next generationAI Compiler, designed to accelerate the training speed of artificial intelligence models. According to Lightning AI, the compiler can increase the training speed of large language models (LLMs) by up to 40% compared to unoptimized code in actual scenarios. This good news has brought new technological breakthroughs to the field of AI.
The release of the "Thunder" compiler marks a big step forward for Lightning AI on the PyTorch open source machine learning framework. The company said that by leveraging multiple GPUs, this new tool aims to improve the efficiency of model training and provide developers with a faster training experience. At the same time, "Thunder" adopts an open source model and follows the Apache 2.0 license, providing users with the opportunity to use it for free, which means that more developers can take advantage of its powerful features.
Lightning AI, previously known as Grid AI, began working with Nvidia as early as 2022 to create the next generation of deep learning technology for PyTorch. Their goal is to provide solutions compatible with Nvidia's product line, including torch.compile, nvFuser, Apex, CUDA Deep Neural Network Library (cuDNN), and OpenAI's Triton.
In addition to technical breakthroughs, Lightning AI also focuses on team building. They invited Thomas Viehmann, a core developer of PyTorch, to lead the "Thunder" compiler project to ensure that it can support generative AI models in multi-GPU environments. This move reflects Lightning AI's emphasis on technological innovation and talent training.
Thunder is now available, following the company’s release of Lightning 2.2 in February. Lightning Studios’ products can be purchased at four pricing tiers: Free for individual developers; Professional for engineers, researchers, and scientists; Team for startups and teams; and Enterprise for large organizations.
Product entry:https://www.1ai.net/6597.html