> 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-V3

[DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) is a large-scale Mixture-of-Experts model with 671B total parameters and 37B activated per token. It features Multi-head Latent Attention (MLA), innovative load balancing, and Multi-Token Prediction (MTP). DeepSeek-V3.2 is an updated release with further improvements.

[Moonlight](https://huggingface.co/moonshotai/Moonlight-16B-A3B) by Moonshot AI also uses this architecture with 16B total / 3B activated parameters.

|                  |                                                    |
| ---------------- | -------------------------------------------------- |
| **Task**         | Text Generation (MoE)                              |
| **Architecture** | `DeepseekV3ForCausalLM` / `DeepseekV32ForCausalLM` |
| **Parameters**   | 671B total / 37B active                            |
| **HF Org**       | [deepseek-ai](https://huggingface.co/deepseek-ai)  |

## Available Models

* **DeepSeek-V3**: 671B total, 37B activated
* **DeepSeek-V3.2** (`DeepseekV32ForCausalLM`): updated architecture
* **Moonlight-16B-A3B** (Moonshot AI): 16B total, 3B activated

## Architectures

* `DeepseekV3ForCausalLM`
* `DeepseekV32ForCausalLM`

## Example HF Models

| Model             | HF ID                                                                                 |
| ----------------- | ------------------------------------------------------------------------------------- |
| DeepSeek-V3       | [`deepseek-ai/DeepSeek-V3`](https://huggingface.co/deepseek-ai/DeepSeek-V3)           |
| DeepSeek-V3-Base  | [`deepseek-ai/DeepSeek-V3-Base`](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base) |
| DeepSeek-V3.2     | [`deepseek-ai/DeepSeek-V3.2`](https://huggingface.co/deepseek-ai/DeepSeek-V3.2)       |
| Moonlight 16B A3B | [`moonshotai/Moonlight-16B-A3B`](https://huggingface.co/moonshotai/Moonlight-16B-A3B) |

## Example Recipes

| Recipe                                                                                                                                                                | Description                                                |
| --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------- |
| [deepseek\_v32\_hellaswag\_pp.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml)             | SFT — DeepSeek-V3.2 on HellaSwag with pipeline parallelism |
| [moonlight\_16b\_te.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/llm_finetune/moonlight/moonlight_16b_te.yaml)                                   | SFT — Moonlight 16B with Transformer Engine                |
| [moonlight\_16b\_te\_packed\_sequence.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/llm_finetune/moonlight/moonlight_16b_te_packed_sequence.yaml) | SFT — Moonlight 16B with packed sequences                  |

## Try with NeMo AutoModel

**1. Install** ([full instructions](/get-started/installation)):

```bash
pip install nemo-automodel
```

**2. Clone the repo** to get the example recipes:

```bash
git clone https://github.com/NVIDIA-NeMo/Automodel.git
cd Automodel
```

This recipe was validated on **32 nodes × 8 GPUs (256 H100s)**. See the [Launcher Guide](/job-launchers/slurm-cluster) for multi-node setup.

**3. Run the recipe** from inside the repo:

```bash
automodel --nproc-per-node=8 examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml
```

**1. Pull the container** and mount a checkpoint directory:

```bash
docker run --gpus all -it --rm \
  --shm-size=8g \
  -v $(pwd)/checkpoints:/opt/Automodel/checkpoints \
  nvcr.io/nvidia/nemo-automodel:26.04.00
```

**2.** Navigate to the AutoModel directory (where the recipes are):

```bash
cd /opt/Automodel
```

**3. Run the recipe**:

```bash
automodel --nproc-per-node=8 examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml
```

See the [Installation Guide](/get-started/installation) and [LLM Fine-Tuning Guide](/recipes-e2e-examples/sft-peft).

## Fine-Tuning

See the [LLM Fine-Tuning Guide](/recipes-e2e-examples/sft-peft) and the [Large MoE Fine-Tuning Guide](/recipes-e2e-examples/large-moe-fine-tuning).

## Hugging Face Model Cards

* [deepseek-ai/DeepSeek-V3](https://huggingface.co/deepseek-ai/DeepSeek-V3)
* [deepseek-ai/DeepSeek-V3-Base](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base)
* [moonshotai/Moonlight-16B-A3B](https://huggingface.co/moonshotai/Moonlight-16B-A3B)