Article
Masked MTP drafts
Avifenesh
• • 1
Avi Fenesh
Avifenesh
AI & ML interests
None yet
Recent Activity
liked a model about 7 hours ago
incoai/GLM-5.3-Flash-DFlash2 repliedto their post about 8 hours ago
Wrote this up because it still feels backwards. Acceptance dropped, tok/s went up.
Full draft head: 66.7% / 117.1 tok/s. Trimmed: 63.6% / 121.7. Later remeasure +5.1% on a Pro 6000, +6.4% on a 5090 laptop. Full head still accepts more. Loses anyway.
Not my idea. FR-Spec. I just wired it for the native MTP head.
https://huggingface.co/blog/Avifenesh/masked-mtp-drafts
https://huggingface.co/Avifenesh/Qwen3.8-27B-NVFP4-MTP-GGUF updated a model about 12 hours ago
tiyuvta/Qwen3.8-Flash-Next-NVFP4Organizations
published an article 16 days ago