88Z AI, 11 August — Meta has released Muse Glimmer, an open-source large language model with 30 billion parameters. Optimized for local deployment, it uses 4-bit quantization and DFlash speculative decoding to run smoothly on devices with 24GB VRAM, such as Macs or consumer GPUs, supporting offline multimodal perception, tool use, and code generation.
At a glance
- Muse Glimmer is a 30-billion parameter open-source model optimized for local execution
- Runs smoothly on 24GB VRAM devices with offline multimodal and coding capabilities
The update
Meta has officially open-sourced Muse Glimmer, a large language model featuring 30 billion parameters. The core design philosophy of this model is to serve local persistent operation scenarios, aiming to lower the barrier to entry for high-performance AI applications. By incorporating 4-bit quantization technology and DFlash speculative decoding algorithms, Muse Glimmer significantly improves inference efficiency, allowing it to run smoothly on devices equipped with 24GB of VRAM, which includes most Mac computers and mainstream consumer-grade GPU hardware. In terms of functionality, the model not only supports independent operation in fully offline environments but also possesses multimodal perception capabilities, enabling it to understand and process various types of data inputs. Furthermore, Muse Glimmer supports external tool invocation and automatic code writing, providing powerful localized AI assistance for developers and individual users, allowing them to execute complex tasks without relying on cloud services.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- APPSOSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.