DiffusionGemma#
DiffusionGemma is a block-diffusion language model from Google. Instead of generating tokens left-to-right, it denoises a fixed-length canvas of tokens in parallel: a causal encoder reads the prompt and a bidirectional decoder iteratively refines the response canvas. The released checkpoint is a Mixture-of-Experts model with 26B total parameters and ~4B active per token.
Task |
Text Generation (Block Diffusion, MoE) |
Architecture |
|
Parameters |
26B total / ~4B active |
HF Org |
Available Models#
DiffusionGemma 26B-A4B-it (
DiffusionGemmaForBlockDiffusion): instruction-tuned block-diffusion MoE.
Architectures#
DiffusionGemmaForBlockDiffusion— block-diffusion MoE (causal prompt encoder + bidirectional canvas decoder).
Example HF Models#
Model |
HF ID |
|---|---|
DiffusionGemma 26B-A4B-it |
Example Recipes#
Recipe |
Description |
|---|---|
Full SFT — DiffusionGemma 26B-A4B with FSDP2 + Expert Parallelism |
|
LoRA SFT — DiffusionGemma 26B-A4B with FSDP2 + Expert Parallelism |
Try with NeMo AutoModel#
1. Install (full instructions):
pip install nemo-automodel
2. Clone the repo to get the example recipes:
git clone https://github.com/NVIDIA-NeMo/Automodel.git
cd Automodel
Note
This recipe was validated with Expert Parallelism (EP=8) on a single 8×H100 node. See the Launcher Guide for multi-node setup.
3. Run the recipe from inside the repo:
automodel --nproc-per-node=8 examples/dllm_sft/diffusion_gemma_sft.yaml
Fine-Tuning#
See the DiffusionGemma Fine-Tuning Guide for the block-diffusion training objective, self-conditioning, and the full list of supported features (SFT, LoRA, Expert Parallelism, activation checkpointing).