Llama-v3.2-3B-Instruct-SSD

Llama-v3.2-3B-Instruct-SSD: Optimized for Qualcomm Devices

Llama 3 is a family of LLMs. The model is quantized to w4a16 (4-bit weights and 16-bit activations) and part of the model is quantized to w8a16 (8-bit weights and 16-bit activations) making it suitable for on-device deployment. This variant uses Self Speculative Decoding (SSD), inference acceleration solution that achieves on-target speed up with guaranteed output accuracy identical to the base model. Choose this variant over llama_v3_2_3b_instruct for faster token generation on supported devices.

This is based on the implementation of Llama-v3.2-3B-Instruct-SSD 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 Llama-v3.2-3B-Instruct-SSD on-device

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

You'll need to export the model artifact using the steps below, then follow Run a Local Model with GenieX.

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

Getting Started

Due to licensing restrictions, we cannot distribute pre-exported model assets for this model. 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

See our repository for Llama-v3.2-3B-Instruct-SSD on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

• Response Rate: Rate of response generation after the first response token.
• Self Speculative Decoding (SSD): Single-model LLM inference acceleration solution that achieves on-target speed up with guaranteed output accuracy identical to the base model. Choose this variant over llama_v3_2_3b_instruct for faster token generation on supported devices.
• Supported languages: English.
• 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).

Performance Summary

License

• The license for the original implementation of Llama-v3.2-3B-Instruct-SSD can be found here.

References

• LLaMA: Open and Efficient Foundation Language Models
• 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