Public DFlash2 draft models for speculative decoding from RadixArk.
AI & ML interests
We are building open source! SGLang: https://www.sglang.io/ Miles: https://github.com/radixark/miles
Recent Activity
-
RadixArk/Muse-Glimmer-NVFP4
Image-Text-to-Text • 18B • Updated • 329k • 10 -
RadixArk/Muse-Glimmer-q4-MLX
Text Generation • 28B • Updated • 1.79k • 12 -
RadixArk/Muse-Glimmer-q4k-dynamic-MLX
Text Generation • 28B • Updated • 339 • 3 -
RadixArk/Muse-Glimmer-q4km-gs128-MLX
Text Generation • 28B • Updated • 266 • 2
Public DFlash2 draft models for speculative decoding from RadixArk.
-
RadixArk/Muse-Glimmer-NVFP4
Image-Text-to-Text • 18B • Updated • 329k • 10 -
RadixArk/Muse-Glimmer-q4-MLX
Text Generation • 28B • Updated • 1.79k • 12 -
RadixArk/Muse-Glimmer-q4k-dynamic-MLX
Text Generation • 28B • Updated • 339 • 3 -
RadixArk/Muse-Glimmer-q4km-gs128-MLX
Text Generation • 28B • Updated • 266 • 2
Qwen3.8 in NVFP4 (W4A4) — production checkpoints plus DSpark drafts for speculative decoding, built for SGLang. Recipes: https://cookbook.sglang.ai