Fast-dLLM
What it is
A family of acceleration techniques for diffusion-based Large Language Models (dLLMs), Vision-Language Models (dVLMs), and Vision-Language-Action (VLA) models. It implements techniques like KV Caching, Parallel Decoding, and Block Diffusion to significantly speed up inference and training.
The problem it solves
High inference latency and computational costs associated with diffusion-based Large Language Models and Vision-Language Models.
What you could use it for
- Accelerating inference for text-based diffusion models (Fast-dLLM v1/v2)
- Speeding up Vision-Language Model (VLM) inference (Fast_dVLM)
- Real-time autonomous driving actions (Fast-dDrive)
- Efficient block-diffusion training and inference
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.