Qwen 3.8 Flash via DwarfStar (twitter.com)

Qwen 3.8 Flash has been integrated into DwarfStar, bringing enhanced performance capabilities specifically for 64GB Mac systems. This update boasts impressive inference speeds ranging from 50 to 70 tokens per second (t/s) and a remarkable prefill rate exceeding 1400 t/s, making it a significant advancement for developers and researchers utilizing AI models on macOS. Currently, the feature is exclusive to Metal, Apple's graphics technology, which leverages the hardware's capabilities for optimized processing. The integration of N-grams on SSD storage, similar to what was introduced in the previous DS4.1F version, adds further efficiency to data handling and retrieval, streamlining workflows for AI applications. This new offering showcases the commitment to improving processing speeds and expands the potential of AI tools on Mac systems, marking a pivotal moment for machine learning practitioners who require high-performance solutions. Thanks to contributions from developer @ivanfioravanti, the update not only enhances capabilities but also sets a new benchmark for inference performance in the AI/ML community.