ZICQ
中 Log in / Sign up
Newsroom LLMs & Foundation Models #Ollama #Gemma 4 #MLX #Multi-Token Prediction #Apple Silicon

Ollama 0.31 Released: Gemma 4 Achieves Up to 90% Performance Boost with MLX Multi-Token Prediction

Avatar of Mr.Xu

By Mr.Xu Compiled & Reviewed by Editorial

Published: · 3 views

中文阅读 (Chinese) English Version
Co-reading Select any sentence in the article to highlight, annotate, or copy a quote card. Join discussion ↓

Summary:The Ollama 0.31 release integrates MLX engine to enable multi-token prediction (MTP) for the Gemma 4 model on Apple Silicon, resulting in a significant performance boost. Performance increased by up to 90% when used with coding agents, as measured by the Aider polyglot benchmark. This update highlights the potential of AI models to run efficiently on local hardware, providing developers with more powerful tools.


Technical Highlights

  1. Multi-Token Prediction (MTP) Technology: Ollama 0.31 integrates MTP through the MLX engine for the Gemma 4 model, enabling more efficient prediction of multiple tokens when handling complex tasks, thereby boosting overall performance.

  2. Performance Boost: Performance increased by up to 90% when used with coding agents, as measured by the Aider polyglot benchmark. This means users can experience faster response times and higher efficiency when running AI models on local devices.

  3. Apple Silicon Optimization: The update specifically optimizes for Apple Silicon devices, taking full advantage of the hardware acceleration capabilities of Apple chips to provide a smoother AI experience for Mac users.

Industry Impact and Developer Recommendations

  • Enhanced Developer Tools: For developers using the Gemma 4 model, the Ollama 0.31 release means they can run models more efficiently on local devices, reducing reliance on cloud resources.

  • Expanded AI Application Scenarios: The performance boost makes AI models more practical in real-time interaction, programming assistance, and other scenarios requiring quick responses. Developers can explore more possibilities for local AI applications.

  • Hardware Optimization Trend: This update also reflects the demand for AI model hardware optimization. Future AI tools may increasingly focus on deep integration with specific hardware architectures to achieve higher performance.

Conclusion

The Ollama 0.31 release marks a significant advancement in the performance optimization of AI models on local devices. Through the MLX engine and multi-token prediction technology, Gemma 4's performance on Apple Silicon has been significantly enhanced, providing developers with more powerful tools and driving the further expansion of AI application scenarios.


Source: Ollama Blog (2026-06-29)

— END —

Tags: #Ollama #Gemma 4 #MLX #Multi-Token Prediction #Apple Silicon

Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.

What's your take on this story?
🚀 Explore & Submit Works

Community Comments

Loading live comments and annotations…