> 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.

# DeepSeek-V4-Flash-Vision-Exp

> Use NeMo AutoModel to fine-tune DeepSeek-V4-Flash-Vision-Exp for VLM workflows with checked-in image-text-to-text recipes and documented distributed settings.

[DeepSeek-V4-Flash-Vision-Exp](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) has a checked-in NeMo AutoModel recipe for image-text-to-text.
The Hugging Face configuration declares the `DeepseekV4ForCausalLM` architecture.

## Fine-Tune DeepSeek-V4-Flash-Vision-Exp

Follow the [installation instructions](/get-started/installation), then review the checked-in recipe and
its declared topology below before choosing a local or cluster launcher.

Use the [Slurm launcher guide](/job-launchers/slurm-cluster) for the multi-node run.

## Choose a Workflow

| Goal                                  | Start Here                                                                                                                                                                     |
| ------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Run the primary recipe for this model | Use the [recipe configuration](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/deepseek_v4/deepseek_v4_flash_vision_medpix_4k_ep32pp4_100steps.yaml). |
| Prepare your environment              | Follow the [installation instructions](/get-started/installation).                                                                                                             |

### Configuration

| Setting  | Configuration                                                                             |
| -------- | ----------------------------------------------------------------------------------------- |
| Hardware | -                                                                                         |
| Strategy | FSDP2; tp\_size=1, pp\_size=4, cp\_size=1, ep\_size=32                                    |
| Nodes    | 16                                                                                        |
| Features | Activation checkpointing (activation\_checkpointing=true), HybridEP (dispatcher=hybridep) |
| Advanced | bfloat16, attention=tilelang, linear=torch, experts=torch\_mm, dispatcher=hybridep        |

## Model Reference

### Model Architecture

| Property     | Value                   |
| ------------ | ----------------------- |
| Task         | Image-text-to-text      |
| Architecture | `DeepseekV4ForCausalLM` |

### Available Models

* [`deepseek-ai/DeepSeek-V4-Flash-Vision-Exp`](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp)

## Related Resources

* [Vision-language model fine-tuning guide](/recipes-e2e-examples/gemma-3-3n)