Qwen2.5-VL-7B-Instruct: Optimized for Qualcomm Devices

Qwen2.5-VL-7B-Instruct is a multimodal vision-language model with 7 billion parameters that can process both images and text, enabling visual question answering, image description, and other vision-language tasks.

This is based on the implementation of Qwen2.5-VL-7B-Instruct found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Deploying Qwen2.5-VL-7B-Instruct on-device

Follow the GenieX quickstart to install GenieX and deploy the model on a target device.

See the LLM-on-Genie tutorial to run with the Genie runtime. Note: Genie support will be deprecated soon.

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
GENIE	w4a16	Snapdragon® 8 Elite Gen 5 Mobile	QAIRT 2.45	Download
GENIE	w4a16	Snapdragon® 8 Elite Mobile	QAIRT 2.45	Download
GENIE	w4a16	Snapdragon® X2 Elite	QAIRT 2.45	Download
GENIE	w4a16	Snapdragon® X Elite	QAIRT 2.45	Download
GENIE	w4a16	Qualcomm® SA8775P	QAIRT 2.45	Download
GENIE	w4a16	Qualcomm® Dragonwing™ IQ-9075	QAIRT 2.45	Download
GENIEX_LLAMACPP	q4_0	Universal		Download
GENIEX_QAIRT	w4a16	Snapdragon® 8 Elite Gen 5 Mobile	QAIRT 2.45	Download
GENIEX_QAIRT	w4a16	Snapdragon® 8 Elite Mobile	QAIRT 2.45	Download
GENIEX_QAIRT	w4a16	Snapdragon® X2 Elite	QAIRT 2.45	Download
GENIEX_QAIRT	w4a16	Snapdragon® X Elite	QAIRT 2.45	Download
GENIEX_QAIRT	w4a16	Qualcomm® Dragonwing™ IQ-9075	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit Qwen2.5-VL-7B-Instruct on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models 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 Qwen2.5-VL-7B-Instruct on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

Supported languages: English, Chinese, and many others.
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. The lower bound is for a short prompt (up to 128 tokens, i.e., one iteration of the prompt processor) and the upper bound is for a prompt using the full context length.
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)
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Snapdragon® 8 Elite Gen 5 Mobile	2048	16.56428756713867	0.0834975 - 1.33596
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Snapdragon® 8 Elite Mobile	2048	15.609390258789062	0.100003 - 1.600048
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Snapdragon® X2 Elite	2048	23.43951416015625	0.0756 - 1.2096
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Snapdragon® X Elite	2048	4.930407524108887	0.162986 - 2.607776
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Qualcomm® Dragonwing™ IQ-9075	2048	10.637053489685059	0.13849799999999998 - 2.2159679999999997
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Qualcomm® Dragonwing™ Q-8750	2048	14.676631736755372	0.1007981 - 1.6127696
Qwen2.5-VL-7B-Instruct	GENIE	w4a16	Qualcomm® Dragonwing™ IQ-X7181	2048	6.205299758911133	0.1729066 - 2.7665056
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	512	14.149997	1.79170775 - 7.166831
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	512	13.935798	1.849144 - 7.396576
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	512	12.359337	1.30130525 - 5.205221
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	4096	8.381849	3.67377984375 - 117.560955
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	4096	8.467235	3.7547086875 - 120.150678
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	4096	10.037572	1.82992665625 - 58.557653
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	512	20.779889	2.2249095000000003 - 8.899638000000001
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	512	20.341266	1.1287639999999999 - 4.5150559999999995
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	512	13.502259	0.7596047499999999 - 3.0384189999999998
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	4096	16.130136	2.0524877499999996 - 65.67960799999999
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	4096	16.363986	1.8921905625 - 60.550098
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	4096	11.546512	0.79908971875 - 25.570871
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X Elite	512	17.148109	0.802278 - 3.209112
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X Elite	512	16.84612	0.8744725 - 3.49789
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X Elite	512	12.056341	1.2022325 - 4.80893
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X Elite	4096	12.229055	1.2957100312499998 - 41.462720999999995
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X Elite	4096	8.527917	1.3487290625000001 - 43.159330000000004
Qwen2.5-VL-7B-Instruct	GENIEX_LLAMACPP	q4_0	Snapdragon® X Elite	4096	10.006898	1.3890724062500002 - 44.450317000000005
Qwen2.5-VL-7B-Instruct	GENIEX_QAIRT	w4a16	Snapdragon® 8 Elite Gen 5 Mobile	2048	17.347405	0.07970000000000001 - 1.2752000000000001
Qwen2.5-VL-7B-Instruct	GENIEX_QAIRT	w4a16	Snapdragon® 8 Elite Mobile	2048	15.12732	0.104 - 1.664
Qwen2.5-VL-7B-Instruct	GENIEX_QAIRT	w4a16	Snapdragon® X2 Elite	2048	23.849203	0.0722 - 1.1552
Qwen2.5-VL-7B-Instruct	GENIEX_QAIRT	w4a16	Snapdragon® X Elite	2048	12.530116	0.1497 - 2.3952
Qwen2.5-VL-7B-Instruct	GENIEX_QAIRT	w4a16	Qualcomm® Dragonwing™ IQ-9075	2048	10.808642	0.1452 - 2.3232

License

The license for the original implementation of Qwen2.5-VL-7B-Instruct can be found here.

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.

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

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/Qwen2.5-VL-7B-Instruct

Qwen2.5-VL Technical Report

Paper • 2502.13923 • Published Feb 19, 2025 • 218