AxForge
Sign in Get started

Model reference · open weights

LLaDA-V

LLMs GSAI-ML Vision + text 1 build Licence not stated 14k dl/mo

LLaDA-V is an open-weight language model from GSAI-ML. LLaDA-V (FP16) weighs 16.9 GB; the smallest configuration that runs it is 2× RTX 3060 12 GB.

Request Rent a GPU

For self-hosting, compare this build's requirements below with the GPU you plan to rent.

  • LLaDA-V is an 8.4B parameter diffusion-based vision-language model developed by GSAI-ML for image-text-to-text tasks.
  • It supports a context length of 16,384 tokens and was introduced in the paper "LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning.

Summary of the GSAI-ML/LLaDA-V model card, 2026-10-01

Hugging Face ↗Paper ↗

What it is

Released byGSAI-ML
Released2025-05-28
Parameters8.4B
VRAM16.9 GB for the weights

What it runs on

Memory and cards for LLaDA-V (FP16)

16.9 GBweights, file size
524 MBcache per 1K tokens
727 MBruntime overhead, at least
16,384 tokenscontext max
CardRequests at onceContext maxMemory
8K each
RTX 3060 12 GB … RTX 4060 Ti 16 GB
2 smaller cards
RTX 3090 24 GB110K23.4 GB
RTX 4090 24 GB110K23.4 GB
RTX 5090 32 GB3all 16K31.0 GB
L40S 48 GB6all 16K44.0 GB
A100 80 GB14all 16K78.2 GB
H100 80 GB13all 16K78.1 GB
RTX PRO 6000 Blackwell 96 GB16all 16K93.8 GB
DGX Spark (GB10) 128 GB unified19all 16K107 GB
H200 141 GB27all 16K138 GB
B200 180 GB35all 16K176 GB
2× RTX 3060 12 GB
tensor parallel		19K11.6 GB a card
2× RTX 4060 Ti 16 GB
tensor parallel		2all 16K15.4 GB a card
2× RTX 5090 32 GB
tensor parallel		10all 16K31.0 GB a card
2× L40S 48 GB
tensor parallel		16all 16K44.0 GB a card
4× RTX 4090 24 GB
tensor parallel		17all 16K23.4 GB a card
4× RTX 3090 24 GB
tensor parallel		17all 16K23.4 GB a card
Memory needed at each load
Requests at once8K tokens each
121.9 GB
539.1 GB
852.0 GB
1686.3 GB
32155 GB
64292 GB

One card, with vLLM's small-card settings.

From the model card

What GSAI-ML says about LLaDA-V

Read the model card

Update (March 2026): We are excited to introduce LLaDA-o, the latest model in the LLaDA series. As an effective and length-adaptive omni diffusion model for unified multimodal understanding and generation, LLaDA-o extends the LLaDA line to broader multimodal settings, supporting visual understanding, text-to-image generation, and instruction-based image editing. For more details, please check out the paper and code.

LLaDA-V

We introduce LLaDA-V, a competitive diffusion-based vision-language model that outperforms other diffusion MLLMs.

It was presented in the paper LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning.

Project Page: https://ml-gsai.github.io/LLaDA-V-demo/

Code: https://github.com/ML-GSAI/LLaDA-V

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms