> 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.

# Qwen3-VL / Qwen3-VL-MoE

[Qwen3-VL](https://qwenlm.github.io/blog/qwen3/) is Alibaba Cloud's third-generation vision language model series. The MoE variant activates a fraction of parameters per token for efficient large-scale inference.

|                  |                                     |
| ---------------- | ----------------------------------- |
| **Task**         | Image-Text-to-Text                  |
| **Architecture** | `Qwen3VLForConditionalGeneration`   |
| **Parameters**   | 4B – 235B                           |
| **HF Org**       | [Qwen](https://huggingface.co/Qwen) |

## Available Models

* **Qwen3-VL-8B-Instruct**: 8B
* **Qwen3-VL-4B-Instruct**: 4B
* **Qwen3-VL-MoE-30B**: 30B total (MoE)
* **Qwen3-VL-MoE-235B**: 235B total (MoE)

## Architecture

* `Qwen3VLForConditionalGeneration`

## Example HF Models

| Model                | HF ID                                                                           |
| -------------------- | ------------------------------------------------------------------------------- |
| Qwen3-VL 4B Instruct | [`Qwen/Qwen3-VL-4B-Instruct`](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) |
| Qwen3-VL 8B Instruct | [`Qwen/Qwen3-VL-8B-Instruct`](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) |

## Example Recipes

| Recipe                                                                                                                                                                  | Dataset    | Description                                                |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ---------------------------------------------------------- |
| [qwen3\_vl\_4b\_instruct\_rdr.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/qwen3/qwen3_vl_4b_instruct_rdr.yaml)                       | rdr-items  | SFT — Qwen3-VL 4B on RDR Items                             |
| [qwen3\_vl\_8b\_instruct\_rdr.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/qwen3/qwen3_vl_8b_instruct_rdr.yaml)                       | rdr-items  | SFT — Qwen3-VL 8B on RDR Items                             |
| [qwen3\_vl\_8b\_cp2\_vision\_frame\_shard.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/qwen3/qwen3_vl_8b_cp2_vision_frame_shard.yaml) | rdr-items  | SFT — dense Qwen3-VL 8B with CP2 and vision frame sharding |
| [qwen3\_vl\_moe\_30b\_te\_deepep.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_te_deepep.yaml)                  | MedPix-VQA | SFT — Qwen3-VL-MoE 30B with TE + DeepEP                    |
| [qwen3\_vl\_moe\_235b.yaml](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/qwen3/qwen3_vl_moe_235b.yaml)                                      | MedPix-VQA | SFT — Qwen3-VL-MoE 235B                                    |

## Try with NeMo AutoModel

**1. Clone and install from source** ([full instructions](/get-started/installation)):

```bash
git clone https://github.com/NVIDIA-NeMo/Automodel.git
cd Automodel
uv sync --locked --all-groups --all-extras --extra vlm-media
```

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

```bash
uv run automodel --nproc-per-node=8 examples/vlm_finetune/qwen3/qwen3_vl_4b_instruct_rdr.yaml
```

#### Run with Docker

**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:26.06.00
```

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

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

**3. Install media dependencies** (not installed in the container by default):

```bash
uv pip install ".[vlm-media]"
```

**4. Run the recipe**:

```bash
automodel --nproc-per-node=8 examples/vlm_finetune/qwen3/qwen3_vl_4b_instruct_rdr.yaml
```

See the [Installation Guide](/get-started/installation) and [VLM Fine-Tuning Guide](/recipes-e2e-examples/gemma-3-3n).

## Fine-Tuning

See the [VLM Fine-Tuning Guide](/recipes-e2e-examples/gemma-3-3n).

Dense Qwen3-VL supports context parallelism. To distribute the vision tower across the CP
group, enable `distributed.multimodal.vision.frame_sharding`; see the
[Context-Parallel Vision Frame Sharding guide](/development/cp-vision-frame-sharding). Qwen3-VL-MoE
does not yet support this path.

## Hugging Face Model Cards

* [Qwen/Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct)
* [Qwen/Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct)