The story’s score and claims

Comments on No. 60 of 123 · blog.comfy.org · edition of Saturday 14 November 2026

Doubling LLM token rates in ComfyUI

Summary

ComfyUI v0.45.0 reports roughly doubled text generation speed for Qwen3.x LLMs, with no change to model or output quality. The gains come from rewritten GEMV and tensor-core kernels for the skinny matrix multiplications that dominate single-sequence decoding, from a hybrid-attention decode kernel, and from an adaptive tree-based multi-token prediction scheme.…

Comments

Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules