DFlash
github.com/z-lab/dflash- Category
- AI Tools
- Pricing
- Open Source
- Type
- TOOL
- Builder
- z-lab
- GitHub
- 5.6k stars
- Added
- Jun 18, 2026
About
DFlash is a lightweight block diffusion model designed for speculative decoding that accelerates LLM inference through efficient parallel token drafting. It provides pre-trained draft models for popular LLMs like Qwen and LLaMA, enabling significant speedups in text generation.
Why it made the leaderboard
A lightweight block diffusion model for speculative decoding: it drafts tokens in parallel to accelerate LLM inference, with pre-trained draft models for Qwen and LLaMA. Works across vLLM, SGLang, Transformers, and MLX, so it slots into an existing serving stack.
Tags
llminferencespeculative-decodingoptimizationtransformersvllmsglang
Tech Stack
Python
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.