README.md

---
library_name: pytorch
license: other
tags:
- llm
- vlm
- generative_ai
- android
pipeline_tag: text-generation

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_vl_2b_instruct/web-assets/model_demo.png)

# Qwen3-VL-2B-Instruct: Optimized for Qualcomm Devices

Qwen3-VL is a vision-language model from Alibaba Cloud capable of understanding both text and images for multimodal reasoning tasks such as visual question answering and image captioning.

This is based on the implementation of Qwen3-VL-2B-Instruct found [here](https://huggingface.co/Qwen/Qwen3-VL-2B-Instruct).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/qwen3_vl_2b_instruct) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.

## Deploying Qwen3-VL-2B-Instruct on-device

Follow the [GenieX quickstart](https://geniex.aihub.qualcomm.com/en/get-started/quickstart) to install GenieX and deploy the model on a target device.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| GENIEX_LLAMACPP | q4_0 | Universal |  | [Download](https://huggingface.co/unsloth/Qwen3-VL-2B-Instruct-GGUF/resolve/main/Qwen3-VL-2B-Instruct-Q4_0.gguf)

For more device-specific assets and performance metrics, visit **[Qwen3-VL-2B-Instruct on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/qwen3_vl_2b_instruct)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/qwen3_vl_2b_instruct) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [Qwen3-VL-2B-Instruct on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/qwen3_vl_2b_instruct) for usage instructions.

## Model Details

**Model Type:** Model_use_case.text_generation

**Model Stats:**
- Model architecture: Transformer with ViT Vision Encoder, Grouped Query Attention (GQA), and SwiGLU activation.
- Supported languages: 100+ languages and dialects
- TTFT: Time To First Token is the time it takes to generate the first response token. This is expressed as a range because it varies based on the length of the prompt.
- Response Rate: Rate of response generation after the first response token.

## Performance Summary
| Model | Runtime | Precision | Chipset | Context Length | Response Rate (tokens per second) | Time To First Token (range, seconds)
|---|---|---|---|---|---|---
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 512 | 41.543766 | 0.46449075 - 1.857963
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 512 | 42.711314 | 0.46647375 - 1.865895
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 512 | 34.032126 | 0.11172275000000001 - 0.44689100000000004
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 512 | 62.900994 | 0.1141745 - 0.456698
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 512 | 64.127228 | 0.112751 - 0.451004
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 512 | 23.365032 | 0.07905475 - 0.316219
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 4096 | 30.033638 | 0.23735575 - 7.595384
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 4096 | 26.100461 | 0.23745440625 - 7.598541
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 4096 | 20.902282 | 0.108134125 - 3.460292
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 512 | 21.177916 | 0.1795745 - 0.718298
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 512 | 20.103331 | 0.18485975 - 0.739439
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 512 | 24.308468 | 0.16061474999999997 - 0.6424589999999999
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 4096 | 15.733537 | 0.43978756249999995 - 14.073201999999998
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 4096 | 15.966534 | 0.46862746875 - 14.996079
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 4096 | 11.584063 | 0.24593868749999998 - 7.870037999999999
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Qualcomm® Dragonwing™ IQ-9075 | 512 | 1.882364 | 1.1826937499999999 - 4.7307749999999995
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Qualcomm® Dragonwing™ IQ-9075 | 512 | 1.878065 | 1.18503875 - 4.740155
| Qwen3-VL-2B-Instruct | GENIEX_LLAMACPP | q4_0 | Qualcomm® Dragonwing™ IQ-9075 | 512 | 19.881743 | 0.18957975 - 0.758319

## License
* The license for the original implementation of Qwen3-VL-2B-Instruct can be found
  [here](https://www.apache.org/licenses/LICENSE-2.0).

## References
* [Qwen3 Technical Report](https://arxiv.org/abs/2505.09388)
* [Source Model Implementation](https://huggingface.co/Qwen/Qwen3-VL-2B-Instruct)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).

## Usage and Limitations

This model may not be used for or in connection with any of the following applications:

- Accessing essential private and public services and benefits;
- Administration of justice and democratic processes;
- Assessing or recognizing the emotional state of a person;
- Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
- Education and vocational training;
- Employment and workers management;
- Exploitation of the vulnerabilities of persons resulting in harmful behavior;
- General purpose social scoring;
- Law enforcement;
- Management and operation of critical infrastructure;
- Migration, asylum and border control management;
- Predictive policing;
- Real-time remote biometric identification in public spaces;
- Recommender systems of social media platforms;
- Scraping of facial images (from the internet or otherwise); and/or
- Subliminal manipulation