We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Lobsters·Jun 28·5 min read
MAX models can now run on Apple silicon GPUs
TL;DR
Apple silicon GPUs previously lacked support for running MAX models, limiting their usability for machine learning tasks. With the 26.4 release, many MAX models can now be executed on M1 to M5 Apple silicon GPUs, enhancing their functionality.
✦ Why It Matters
Engineers can now utilize Apple silicon GPUs for efficient execution of various MAX models in their projects.
Key Takeaways
- Explore running MAX models on Apple silicon GPUs for enhanced performance.
- Test compatibility of models on different M-series systems and report issues.
- Utilize M5 systems for optimal execution of machine learning tasks.
- Set up MAX to experiment with text generation and model serving.
Full Summary
Support for MAX models on Apple silicon GPUs has been significantly enhanced, addressing the gap in compatibility for machine learning applications. The 26.4 release enables many MAX models to run on M1 through M5 GPUs, with ongoing improvements in nightly builds.
Specifically, models such as text large language models (LLMs), vision models, and image diffusion models are now operational on these devices. M5 systems are particularly optimized for this task due to their Neural Accelerators, which facilitate dedicated matrix-multiplication operations.
Although some older M-series systems may experience compatibility issues, users are encouraged to report any problems. Preliminary tests indicate fast model execution, although formal benchmarks against other frameworks like MLX are still pending.
This development opens new avenues for engineers and researchers to leverage Apple silicon for advanced machine learning tasks.