> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/automodel/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo/automodel/_mcp/server.

# DFlash

> Supported DFlash draft checkpoint and NeMo AutoModel recipes.

[Qwen3-4B-DFlash-b16](https://huggingface.co/z-lab/Qwen3-4B-DFlash-b16) is a diffusion draft model trained against a frozen Qwen3 target for speculative decoding.

|                  |                                                          |
| ---------------- | -------------------------------------------------------- |
| **Task**         | Diffusion Draft-Model Training                           |
| **Architecture** | `DFlashDraftModel` with Hugging Face `trust_remote_code` |
| **Target**       | `Qwen/Qwen3-4B`                                          |
| **HF Org**       | [z-lab](https://huggingface.co/z-lab)                    |

## Example Model and Recipe

| Model                                                                           | Recipe                                                                                             |
| ------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- |
| [`z-lab/Qwen3-4B-DFlash-b16`](https://huggingface.co/z-lab/Qwen3-4B-DFlash-b16) | [DFlash SFT](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/dllm_sft/dflash_sft.yaml) |

```bash
uv run torchrun --nproc-per-node=8 nemo_automodel/recipes/dllm/train_ft.py \
  -c examples/dllm_sft/dflash_sft.yaml
```

See the [dLLM Fine-Tuning Guide](/recipes-e2e-examples/dllm-fine-tuning).