---
license: apache-2.0
language:
  - en
library_name: transformers
base_model:
  - Qwen/Qwen3-VL-4B-Thinking
pipeline_tag: image-text-to-text
tags:
  - visual-grounding
  - multimodal
  - qwen3-vl
  - supervised-fine-tuning
---

# EGM-Qwen3-VL-4B-SFT

<p align="center">
  <a href="https://nvlabs.github.io/EGM">[Project Page]</a> &nbsp;
  <a href="https://github.com/NVlabs/EGM">[Code]</a> &nbsp;
</p>

<div align="center">
  <img src="https://nvlabs.github.io/EGM/figure4.jpeg" width="90%"/>
</div>

## Model Summary

**EGM-Qwen3-VL-4B-SFT** is the supervised fine-tuning (SFT) checkpoint from the first stage of the [EGM (Efficient Visual Grounding Language Models)](https://nvlabs.github.io/EGM) training pipeline. It is built on top of [Qwen3-VL-4B-Thinking](https://huggingface.co/Qwen/Qwen3-VL-4B-Thinking).

This is an **intermediate checkpoint** intended for further reinforcement learning training. For the final model with best performance, see [nvidia/EGM-4B](https://huggingface.co/nvidia/EGM-4B).

## Training Details

### SFT Stage

In the SFT stage, a proprietary VLM generates detailed chain-of-thought reasoning steps for visual grounding training data. The base Qwen3-VL-4B-Thinking model is then fine-tuned on this reasoning-augmented data to learn structured visual grounding with explicit reasoning.

This SFT checkpoint serves as the initialization for the subsequent RL stage (GRPO), which yields the final [EGM-4B](https://huggingface.co/nvidia/EGM-4B) model.

### How to Use for RL Training

```bash
pip install -U huggingface_hub
huggingface-cli download nvidia/EGM-4B-SFT --local-dir ./models/EGM-4B-SFT
```

Then follow the installation and RL training instructions in the [EGM repository](https://github.com/NVlabs/EGM#rl-training).

## Model Architecture

| Component | Details |
|---|---|
| Architecture | Qwen3VLForConditionalGeneration |
| Precision | bfloat16 |
| Text Hidden Size | 2560 |
| Text Layers | 36 |
| Attention Heads | 32 (8 KV heads) |
| Text Intermediate Size | 9728 |
| Vision Hidden Size | 1024 |
| Vision Layers | 24 |
| Patch Size | 16 x 16 |
| Max Position Embeddings | 262,144 |
| Vocabulary Size | 151,936 |

## Related Models

| Model | Description |
|---|---|
| [nvidia/EGM-4B](https://huggingface.co/nvidia/EGM-4B) | Final RL-trained model (best performance) |
| [nvidia/EGM-8B-SFT](https://huggingface.co/nvidia/EGM-8B-SFT) | SFT checkpoint for the 8B variant |
| [nvidia/EGM-8B](https://huggingface.co/nvidia/EGM-8B) | Final RL-trained 8B model |

## Citation

```bibtex
@article{zhan2026EGM,
    author = {Zhan, Guanqi and Li, Changye and Liu, Zhijian and Lu, Yao and Wu, Yi and Han, Song and Zhu, Ligeng},
    title = {EGM: Efficient Visual Grounding Language Models},
    booktitle = {arXiv},
    year = {2026}
}
```

## Acknowledgment

This repository benefits from [Qwen3-VL](https://github.com/QwenLM/Qwen3-VL), [InternVL](https://github.com/OpenGVLab/InternVL), [verl](https://github.com/volcengine/verl) and [verl-internvl](https://github.com/Weiyun1025/verl-internvl).