Hands-on Tutorial
KDD 2026 Hands-on Tutorial
A practical tutorial on training, inference, evaluation, and extension with dLLM—a unified open-source framework for diffusion language models.
Overview
A unified framework for dLLM training, inference, and evaluation.
Run inference, adapt a model, reproduce results, and add a custom sampler.
Researchers and practitioners building with diffusion language models.
Schedule
Session outcome Run a first open-model dLLM inference and understand how the repository fits together.
Session outcome Compare sampler behavior, output quality, and latency under a controlled prompt.
Session outcome Launch a reusable training recipe and trace each setting to the unified trainer.
Session outcome Reproduce one benchmark result with a clear, comparable evaluation setup.
Session outcome Implement and verify a small sampler or scheduler extension without changing the pipeline.
Session outcome Diagnose common blockers and connect open questions to practical research directions.
Resources
Unified training, inference, and evaluation framework for diffusion language models.
Report dLLM: Simple Diffusion Language ModelingTechnical report introducing the dLLM framework and reproducible recipes.
Models dLLM Hub on Hugging FaceReleased checkpoints and collections including Tiny-A2D and BERT-Chat.
People
In-person presenter
Ph.D. student, University of Illinois Urbana-Champaign
Works on diffusion language models, post-training, open recipes, model adaptation, and unified evaluation workflows.
Contributor
Ph.D. student, University of California, Berkeley
Works on diffusion language models, agentic AI, and reproducible open-source tooling.
Contributor
Professor, University of Illinois Urbana-Champaign
Works on large-scale data mining and machine learning for graph and multimedia data.
Contributor
Professor, University of California, Berkeley
Works on AI safety and security, agentic AI, privacy, and trustworthy AI systems.
