README.md

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

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/phi_4_mini_instruct/web-assets/model_demo.png)

# Phi-4-Mini-Instruct: Optimized for Qualcomm Devices

Phi-4-mini-instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.

This is based on the implementation of Phi-4-Mini-Instruct found [here](https://huggingface.co/microsoft/Phi-4-mini-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/v0.57.1/src/qai_hub_models/models/phi_4_mini_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 Phi-4-Mini-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/bartowski/microsoft_Phi-4-mini-instruct-GGUF/resolve/main/microsoft_Phi-4-mini-instruct-Q4_0.gguf)

For more device-specific assets and performance metrics, visit **[Phi-4-Mini-Instruct on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/phi_4_mini_instruct)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.57.1/src/qai_hub_models/models/phi_4_mini_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 [Phi-4-Mini-Instruct on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.57.1/src/qai_hub_models/models/phi_4_mini_instruct) for usage instructions.

## Model Details

**Model Type:** Model_use_case.text_generation

**Model Stats:**
- Number of parameters: 3.8B
- 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 (4096 tokens).
- 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)
|---|---|---|---|---|---|---
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Gen 5 Mobile | 512 | 23.272858 | 0.92645325 - 3.705813
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Gen 5 Mobile | 512 | 22.664968 | 1.03434575 - 4.137383
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Gen 5 Mobile | 512 | 19.242405 | 0.17659049999999998 - 0.7063619999999999
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Gen 5 Mobile | 4096 | 12.621046 | 2.2515993749999996 - 72.05117999999999
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Gen 5 Mobile | 4096 | 12.03971 | 2.4794271875 - 79.34167
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Gen 5 Mobile | 4096 | 11.220493 | 0.36033615625 - 11.530757
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 512 | 24.055461 | 0.9468802500000001 - 3.7875210000000004
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 512 | 23.939526 | 0.95809125 - 3.832365
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 512 | 19.557942 | 0.1830165 - 0.732066
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 4096 | 13.965758 | 1.94027484375 - 62.088795
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 4096 | 8.177678 | 2.3551633124999998 - 75.36522599999999
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® 8 Elite Mobile | 4096 | 13.116693 | 0.35286768749999997 - 11.291765999999999
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 512 | 33.714527 | 0.24767750000000002 - 0.9907100000000001
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 512 | 33.35471 | 0.248513 - 0.994052
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 512 | 22.484254 | 0.12369325 - 0.494773
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 4096 | 24.554353 | 0.44789846875 - 14.332751
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 4096 | 24.309187 | 0.4479129375 - 14.333214
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X2 Elite | 4096 | 15.507219 | 0.19384584375 - 6.203067
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 512 | 27.988389 | 0.40832025 - 1.633281
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 512 | 27.143556 | 0.46132475 - 1.845299
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 512 | 16.042448 | 0.25184725 - 1.007389
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 4096 | 15.617096 | 0.913791375 - 29.241324
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 4096 | 16.000366 | 0.9590768749999999 - 30.690459999999998
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Snapdragon® X Elite | 4096 | 9.673267 | 0.41174053125 - 13.175697
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Qualcomm® QCS9075 | 4096 | 13.1 | 1.9971918880000001 - 63.91014
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Qualcomm® QCS9075 | 4096 | 3.0 | 2.471042471 - 79.073359
| Phi-4-Mini-Instruct | GENIEX_LLAMACPP | q4_0 | Qualcomm® QCS9075 | 4096 | 10.0 | 0.44137930999999997 - 14.124138

## License
* The license for the original implementation of Phi-4-Mini-Instruct can be found
  [here](https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/main/LICENSE).

## References
* [Phi-4 Technical Report](https://arxiv.org/abs/2412.08905)
* [Source Model Implementation](https://huggingface.co/microsoft/Phi-4-mini-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