llada

[ECCV 2026] A diffusion-based framework for document OCR that replaces autoregressive decoding with block-level parallel diffusion decoding.

Local inference for diffusion language models (LLaDA, Dream), written in Mojo

[ACL '26] UNCODE (a.k.a. PC-Sampler): training-free decoding calibration for masked diffusion LLMs. Code for "Empirical Analysis of Decoding Biases in Masked Diffusion Models"

LLaDA implementation

OpenAI-compatible HTTP serving for diffusion language models. Continuous batching + LocalLeap acceleration.

GUI for LLaDA Diffusion LLM with Quantization for low end GPU and CPU options. Now with Prototype Training and Vector DB.

SPOT: lossless speculative decoding for diffusion language models via commit-order prediction

Minimal quickstart fork of the LLaDA repository. Watch text diffusion with a single line of code (& flag).

Official implementation of Attention-Discounted Adaptive Sampler (ADAS) for masked diffusion language models.

[NeurIPS 2026] Official implementation of "Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs"

[EMNLP 2026] Length-Adaptive Decoding for Masked Diffusion Machine Translation — a training-free target-length selector for masked diffusion LMs.

T²Quant (T2Quant): task- and time-aware activation quantization for diffusion language models (LLaDA, Dream). Label-free calibration + exact DP precision schedule. arXiv:2610.31198

Minimal masked diffusion LM in PyTorch. Pretraining + SFT + a scaling family from 50M to 350M, LLaDA recipe.

Any LLM can serve as SystemOne decision model on your Personal Computer. And support for LLaDA-MoE/DiffusionGemma.

LLaDA based ASR

Reachability, not topology: why showing a diffusion language model the real citation graph seemed to hurt, and the one-mask fix (LLaDA-8B, Cora)

DiffLM Lab — Interactive Diffusion Language Model Laboratory. 6 modules: denoising playground, forward absorbing process, sampling strategies, AR-vs-diffusion race, block diffusion, masked-diffusion ELBO. Real in-browser denoiser.

AR → discrete diffusion LLM backend (LLaDA 2.1 recipe, Qwen3-0.6B, Pascal/sm_61)

This project is an implementation of a character-level diffusion model (LLaDA) for text generation, based on the principles outlined in the take-home exercise. It also includes a standard autoregressive Transformer model as a baseline for comparison.

Pure-CUDA inference engine + OpenAI-compatible server for LLaDA2.2-mini (16B MoE diffusion LLM) on Jetson Orin Nano Super 8GB