Search Results for "100"

...It is specifically optimized for local and resource-constrained deployment and has been validated on NVIDIA DGX Spark, Apple Silicon MacBooks, and Mac mini systems. FP8 testing reached around 100–105 tokens/s on DGX Spark and 86–90 tokens/s on an M4 Pro MacBook. BF16, FP8, and INT4 weights are available, while deployment options include SGLang, vLLM, and experimental Ollama support on Apple Silicon.

...Distilled from the larger Muse Spark, it combines multi-step reasoning, reliable tool use, coding, failure recovery, and image understanding in a dense 29.6B-parameter architecture with a dedicated 1.8B-parameter perception encoder. It supports more than 100 languages and a 131K+ token context window, allowing agents to maintain coherent plans across extended workflows. Muse Glimmer can interpret screenshots, charts, documents, and images alongside text, while configurable reasoning strength lets developers balance response quality and speed. Its quantized variants reduce the model below 20 GB for operation on systems with 24–32 GB of memory, and DFlash speculative decoding can substantially accelerate generation.