v0.57.0
Browse filesSee https://github.com/qualcomm/ai-hub-models/releases/v0.57.0 for changelog.
- LICENSE +1 -0
- README.md +116 -0
- release_assets.json +87 -0
LICENSE
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
The license of the original trained model can be found at https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE.
|
README.md
ADDED
|
@@ -0,0 +1,116 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: pytorch
|
| 3 |
+
license: other
|
| 4 |
+
tags:
|
| 5 |
+
- llm
|
| 6 |
+
- generative_ai
|
| 7 |
+
- android
|
| 8 |
+
pipeline_tag: text-generation
|
| 9 |
+
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+

|
| 13 |
+
|
| 14 |
+
# Qwen3-8B: Optimized for Qualcomm Devices
|
| 15 |
+
|
| 16 |
+
The Qwen3-8B is a state-of-the-art multilingual base language model with 8 billion parameters, excelling in language understanding, generation, coding, and mathematics.
|
| 17 |
+
|
| 18 |
+
This is based on the implementation of Qwen3-8B found [here](https://huggingface.co/Qwen/Qwen3-8B).
|
| 19 |
+
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/v0.57.0/src/qai_hub_models/models/qwen3_8b) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
|
| 20 |
+
|
| 21 |
+
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.
|
| 22 |
+
|
| 23 |
+
## Deploying Qwen3-8B on-device
|
| 24 |
+
|
| 25 |
+
Follow the [GenieX quickstart](https://geniex.aihub.qualcomm.com/en/get-started/quickstart) to install GenieX and deploy the model on a target device.
|
| 26 |
+
|
| 27 |
+
See the [LLM-on-Genie](https://github.com/qualcomm/ai-hub-apps/tree/main/tutorials/llm_on_genie) tutorial to run with the Genie runtime. Note: Genie support will be deprecated soon.
|
| 28 |
+
|
| 29 |
+
## Getting Started
|
| 30 |
+
There are two ways to deploy this model on your device:
|
| 31 |
+
|
| 32 |
+
### Option 1: Download Pre-Exported Models
|
| 33 |
+
|
| 34 |
+
Below are pre-exported model assets ready for deployment.
|
| 35 |
+
|
| 36 |
+
| Runtime | Precision | Chipset | SDK Versions | Download |
|
| 37 |
+
|---|---|---|---|---|
|
| 38 |
+
| GENIE | w4a16 | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_8_elite_gen5.zip)
|
| 39 |
+
| GENIE | w4a16 | Snapdragon® 8 Elite Mobile | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_8_elite.zip)
|
| 40 |
+
| GENIE | w4a16 | Snapdragon® X2 Elite | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_x2_elite.zip)
|
| 41 |
+
| GENIE | w4a16 | Snapdragon® X Elite | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_x_elite.zip)
|
| 42 |
+
| GENIE | w4a16 | Qualcomm® SA8775P | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_sa8775p.zip)
|
| 43 |
+
| GENIE | w4a16 | Qualcomm® QCS9075 | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_qcs9075.zip)
|
| 44 |
+
| GENIEX_QAIRT | w4a16 | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_8_elite_gen5.zip)
|
| 45 |
+
| GENIEX_QAIRT | w4a16 | Snapdragon® 8 Elite Mobile | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_8_elite.zip)
|
| 46 |
+
| GENIEX_QAIRT | w4a16 | Snapdragon® X2 Elite | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_x2_elite.zip)
|
| 47 |
+
| GENIEX_QAIRT | w4a16 | Snapdragon® X Elite | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_x_elite.zip)
|
| 48 |
+
| GENIEX_QAIRT | w4a16 | Qualcomm® QCS9075 | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_qcs9075.zip)
|
| 49 |
+
|
| 50 |
+
For more device-specific assets and performance metrics, visit **[Qwen3-8B on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/qwen3_8b)**.
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
### Option 2: Export with Custom Configurations
|
| 54 |
+
|
| 55 |
+
Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.57.0/src/qai_hub_models/models/qwen3_8b) Python library to compile and export the model with your own:
|
| 56 |
+
- Custom weights (e.g., fine-tuned checkpoints)
|
| 57 |
+
- Custom input shapes
|
| 58 |
+
- Target device and runtime configurations
|
| 59 |
+
|
| 60 |
+
This option is ideal if you need to customize the model beyond the default configuration provided here.
|
| 61 |
+
|
| 62 |
+
See our repository for [Qwen3-8B on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.57.0/src/qai_hub_models/models/qwen3_8b) for usage instructions.
|
| 63 |
+
|
| 64 |
+
## Model Details
|
| 65 |
+
|
| 66 |
+
**Model Type:** Model_use_case.text_generation
|
| 67 |
+
|
| 68 |
+
**Model Stats:**
|
| 69 |
+
- Input sequence length for Prompt Processor: 128
|
| 70 |
+
- Context lengths: 512, 1024, 2048, 3072, 4096
|
| 71 |
+
- Minimum QNN SDK version required: 2.45.0
|
| 72 |
+
- Supported languages: 100+ languages and dialects.
|
| 73 |
+
- 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).
|
| 74 |
+
- Response Rate: Rate of response generation after the first response token. Measured on a short prompt with a long response (with thinking); may slow down when using longer context lengths.
|
| 75 |
+
|
| 76 |
+
## Performance Summary
|
| 77 |
+
| Model | Runtime | Precision | Chipset | Context Length | Response Rate (tokens per second) | Time To First Token (range, seconds)
|
| 78 |
+
|---|---|---|---|---|---|---
|
| 79 |
+
| Qwen3-8B | GENIEX_QAIRT | w4a16 | Snapdragon® 8 Elite Gen 5 Mobile | 4096 | 16.128711 | 0.372668 - 11.925376
|
| 80 |
+
| Qwen3-8B | GENIEX_QAIRT | w4a16 | Snapdragon® 8 Elite Mobile | 4096 | 13.685263 | 0.46723200000000004 - 14.951424000000001
|
| 81 |
+
| Qwen3-8B | GENIEX_QAIRT | w4a16 | Snapdragon® X2 Elite | 4096 | 23.325354 | 0.27114 - 8.67648
|
| 82 |
+
| Qwen3-8B | GENIEX_QAIRT | w4a16 | Snapdragon® X Elite | 4096 | 12.949668 | 0.6188400000000001 - 19.802880000000002
|
| 83 |
+
| Qwen3-8B | GENIEX_QAIRT | w4a16 | Qualcomm® QCS9075 | 4096 | 10.520227 | 0.651428 - 20.845696
|
| 84 |
+
|
| 85 |
+
## License
|
| 86 |
+
* The license for the original implementation of Qwen3-8B can be found
|
| 87 |
+
[here](https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE).
|
| 88 |
+
|
| 89 |
+
## References
|
| 90 |
+
* [Qwen3 Technical Report](https://arxiv.org/abs/2505.09388)
|
| 91 |
+
* [Source Model Implementation](https://huggingface.co/Qwen/Qwen3-8B)
|
| 92 |
+
|
| 93 |
+
## Community
|
| 94 |
+
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
|
| 95 |
+
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
|
| 96 |
+
|
| 97 |
+
## Usage and Limitations
|
| 98 |
+
|
| 99 |
+
This model may not be used for or in connection with any of the following applications:
|
| 100 |
+
|
| 101 |
+
- Accessing essential private and public services and benefits;
|
| 102 |
+
- Administration of justice and democratic processes;
|
| 103 |
+
- Assessing or recognizing the emotional state of a person;
|
| 104 |
+
- Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
|
| 105 |
+
- Education and vocational training;
|
| 106 |
+
- Employment and workers management;
|
| 107 |
+
- Exploitation of the vulnerabilities of persons resulting in harmful behavior;
|
| 108 |
+
- General purpose social scoring;
|
| 109 |
+
- Law enforcement;
|
| 110 |
+
- Management and operation of critical infrastructure;
|
| 111 |
+
- Migration, asylum and border control management;
|
| 112 |
+
- Predictive policing;
|
| 113 |
+
- Real-time remote biometric identification in public spaces;
|
| 114 |
+
- Recommender systems of social media platforms;
|
| 115 |
+
- Scraping of facial images (from the internet or otherwise); and/or
|
| 116 |
+
- Subliminal manipulation
|
release_assets.json
ADDED
|
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "0.57.0",
|
| 3 |
+
"precisions": {
|
| 4 |
+
"w4a16": {
|
| 5 |
+
"chipset_assets": {
|
| 6 |
+
"qualcomm-snapdragon-8-elite": {
|
| 7 |
+
"genie": {
|
| 8 |
+
"tool_versions": {
|
| 9 |
+
"qairt": "2.45.0.260326154327"
|
| 10 |
+
},
|
| 11 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_8_elite.zip"
|
| 12 |
+
},
|
| 13 |
+
"geniex_qairt": {
|
| 14 |
+
"tool_versions": {
|
| 15 |
+
"qairt": "2.45.0.260326154327"
|
| 16 |
+
},
|
| 17 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_8_elite.zip"
|
| 18 |
+
}
|
| 19 |
+
},
|
| 20 |
+
"qualcomm-qcs9075": {
|
| 21 |
+
"genie": {
|
| 22 |
+
"tool_versions": {
|
| 23 |
+
"qairt": "2.45.0.260326154327"
|
| 24 |
+
},
|
| 25 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_qcs9075.zip"
|
| 26 |
+
},
|
| 27 |
+
"geniex_qairt": {
|
| 28 |
+
"tool_versions": {
|
| 29 |
+
"qairt": "2.45.0.260326154327"
|
| 30 |
+
},
|
| 31 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_qcs9075.zip"
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"qualcomm-sa8775p": {
|
| 35 |
+
"genie": {
|
| 36 |
+
"tool_versions": {
|
| 37 |
+
"qairt": "2.45.0.260326154327"
|
| 38 |
+
},
|
| 39 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_sa8775p.zip"
|
| 40 |
+
}
|
| 41 |
+
},
|
| 42 |
+
"qualcomm-snapdragon-8-elite-gen5": {
|
| 43 |
+
"genie": {
|
| 44 |
+
"tool_versions": {
|
| 45 |
+
"qairt": "2.45.0.260326154327"
|
| 46 |
+
},
|
| 47 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_8_elite_gen5.zip"
|
| 48 |
+
},
|
| 49 |
+
"geniex_qairt": {
|
| 50 |
+
"tool_versions": {
|
| 51 |
+
"qairt": "2.45.0.260326154327"
|
| 52 |
+
},
|
| 53 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_8_elite_gen5.zip"
|
| 54 |
+
}
|
| 55 |
+
},
|
| 56 |
+
"qualcomm-snapdragon-x-elite": {
|
| 57 |
+
"genie": {
|
| 58 |
+
"tool_versions": {
|
| 59 |
+
"qairt": "2.45.0.260326154327"
|
| 60 |
+
},
|
| 61 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_x_elite.zip"
|
| 62 |
+
},
|
| 63 |
+
"geniex_qairt": {
|
| 64 |
+
"tool_versions": {
|
| 65 |
+
"qairt": "2.45.0.260326154327"
|
| 66 |
+
},
|
| 67 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_x_elite.zip"
|
| 68 |
+
}
|
| 69 |
+
},
|
| 70 |
+
"qualcomm-snapdragon-x2-elite": {
|
| 71 |
+
"genie": {
|
| 72 |
+
"tool_versions": {
|
| 73 |
+
"qairt": "2.45.0.260326154327"
|
| 74 |
+
},
|
| 75 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-genie-w4a16-qualcomm_snapdragon_x2_elite.zip"
|
| 76 |
+
},
|
| 77 |
+
"geniex_qairt": {
|
| 78 |
+
"tool_versions": {
|
| 79 |
+
"qairt": "2.45.0.260326154327"
|
| 80 |
+
},
|
| 81 |
+
"download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/qwen3_8b/releases/v0.57.0/qwen3_8b-geniex_qairt-w4a16-qualcomm_snapdragon_x2_elite.zip"
|
| 82 |
+
}
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
}
|
| 86 |
+
}
|
| 87 |
+
}
|