Ministral-3-3B-Instruct-2512: Optimized for Qualcomm Devices

Ministral is a compact language model from Mistral AI designed for on-device deployment, offering strong instruction-following capabilities with low latency.

This is based on the implementation of Ministral-3-3B-Instruct-2512 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 Ministral-3-3B-Instruct-2512 on-device

Follow the GenieX 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

For more device-specific assets and performance metrics, visit Ministral-3-3B-Instruct-2512 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 Ministral-3-3B-Instruct-2512 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

• Model architecture: Dense Transformer with Sliding Window Attention and Grouped Query Attention (GQA).
• Supported languages: English, French, German, Spanish, Italian, Portuguese
• 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)
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	512	26.784283	0.9327475 - 3.73099
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	512	25.897794	0.9703127500000001 - 3.8812510000000002
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	512	21.999012	0.190215 - 0.76086
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	4096	12.565397	2.139108875 - 68.451484
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	4096	10.796779	2.5673956875 - 82.156662
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® 8 Elite Mobile	4096	14.63085	0.36086071875 - 11.547543
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	512	39.522372	0.51052025 - 2.042081
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	512	33.041186	0.5579775 - 2.23191
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	512	26.247193	0.130388 - 0.521552
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	4096	25.252779	1.27059640625 - 40.659085
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	4096	25.43421	1.11475046875 - 35.672015
Ministral-3-3B-Instruct-2512	GENIEX_LLAMACPP	q4_0	Snapdragon® X2 Elite	4096	18.565389	0.18276974999999998 - 5.848631999999999

License

• The license for the original implementation of Ministral-3-3B-Instruct-2512 can be found here.

References

• Ministral 3
• Source Model Implementation

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