Inference & Diffusion LLM Resources

Diffusion LLM startups

• Inception Labs — Mercury diffusion LLMs for reasoning, coding, and low-latency applications.

Inference startups

• PrismML — Bonsai low-bit models for efficient on-device inference.
• Inco AI — DFlash team.
• Mirai — On-device inference, Apple Silicon models, uzu inference engine, and quantization.
• diffuspace - on-device dLLM for AI PCs, robotics, intelligent vehicles, and smart homes.

Speculative decoding

• DFlash
• DFlash 2
• DSpark

dLLM

• DiffusionGemma Technical Report — Google DeepMind’s open-weight discrete diffusion model.

Continuous diffusion LLMs

• Plaid: Likelihood-Based Diffusion Language Models
• RePlaid
• Sigma: Large Language Continuous Diffusion Models
• Clock Diffusion: Efficient Semi-Autoregressive Continuous Diffusion Language Models — builds on RePlaid’s training framework.