Dllm

Visit Tool

dLLM is an open-source Coding & Development tool that unifies the training and evaluation of diffusion language models. It provides scalable training pipelines with support for various distributed training methods.

Claim this tool

1View

At a glance

Pricing

Open Source

Free tier

Yes

API

Skill level

Technical

About

What is dllm?

dLLM is an open-source library designed to bring transparency and reproducibility to the development pipeline of diffusion language models. It offers scalable training pipelines, supporting advanced features like LoRA, DeepSpeed, and FSDP, based on the transformers Trainer. The library also provides unified evaluation pipelines built on lm-evaluation-harness, simplifying inference and customization. dLLM includes minimal training, inference, and evaluation recipes for open-weight models such as LLaDA and Dream, and implements various training algorithms like MDLM (masked diffusion) and BD3LM (block diffusion). It also supports accelerated inference and evaluation with Fast-dLLM, offering cache and confidence-threshold decoding.

Best used for

Ideal for developers and AI researchers who need to efficiently train and evaluate diffusion language models, finetune existing models, and implement advanced training algorithms. Especially valuable for those working with open-weight models like LLaDA and Dream, seeking reproducible and scalable development pipelines.

Common actions

train language models

evaluate language models

finetune diffusion models

develop AI chatbots

accelerate model inference

"AI Agents"face swappinggithub copilotworkflowsdeepfakecollaborationopen-sourceautomated workflowlow-code/no-code

Capabilities

Key features

Scalable training pipelines
Unified evaluation pipelines
LoRA, DeepSpeed, FSDP support
Open-weight model recipes
Accelerated inference (Fast-dLLM)
Masked diffusion training
Block diffusion training

Target Audience

developer

Integrations

Not yet documented

Pricing & Plans

Open Source

Free

FAQs

What types of diffusion language models can be trained with dLLM?

dLLM supports training and evaluation for various diffusion language models, including open-weight models like LLaDA and Dream. It also provides recipes for adapting any autoregressive model (e.g., Qwen, LLaMA, GPT-2) into a diffusion model and finetuning BERT encoders into diffusion chatbots.

Does dLLM support distributed training and acceleration?

Yes, dLLM offers scalable training pipelines with support for distributed training methods such as LoRA, DeepSpeed, and FSDP. It also includes Fast-dLLM for accelerated inference and evaluation, featuring cache and confidence-threshold decoding to speed up model performance.

How does dLLM ensure reproducibility in model development?

dLLM unifies training and evaluation pipelines, providing a consistent framework for developing diffusion language models. It offers minimal, open recipes for various models and algorithms, ensuring that the entire development process is transparent and reproducible for researchers and developers.

Trending

Subcategories trending in Coding & Development

Code Assistants DevOps & Infrastructure No-Code / Low-Code Testing & QA Backend & APIs Prompt Engineering

Trending

Also listed in

This tool also appears in

AI Agents & Automation › AI Frameworks & Infra

Explore

Browse AI tools by category

Content & Design Productivity & Business Coding & Development AI Agents & Automation Research & Education Wellness & Lifestyle Career Development Marketing & Growth Data & Analytics Customer Support & CX Finance E-commerce