README.md

Implementing a Chatbot with Llama 3.3 70B on AIC100 Ultra

Table of Contents

• Overview
• Features
• Install Platform and App SDK
• Enable Root Access
• Activate the qeff Environment
• Download the Pre-compiled Llama 3.3 70B Model
• Extract the Model
• Download the code demo.py in this repository
• Demo
• Example

---

1. Overview

This guide outlines the steps to implement a chatbot using the Llama 3.3 70B model on the AIC100 Ultra platform.

---

2. Features

• Pre-compiled Llama3.3 70B models with 8k ctx_len.
• Python API "QEfficient.generation.text_generation_inference.cloud_ai_100_exec_kv".
• Enabling steam function for smooth inference result.
• while loop to avoid reload model to AIC100 Ultra card.

---

3. Install Platform and App SDK

Follow the official installation guide from Efficient Transformers:
👉 https://quic.github.io/efficient-transformers/source/installation.html

4. Enable Root Access

sudo -i

5. Activate the qeff Environment

source /opt/qti-aic/dev/python/qeff/bin/activate

6. Download the Pre-compiled Llama 3.3 70B Model

Download from:
https://qualcom-qpc-models.s3-accelerate.amazonaws.com/SDK1.19.6/meta-llama/Llama-3.3-70B-Instruct/qpc_16cores_128pl_8192cl_1fbs_4devices_mxfp6_mxint8.tar.gz

7. Extract the Model

Extract the tarball to your desired directory:

tar -xzvf qpc_16cores_128pl_8192cl_1fbs_4devices_mxfp6_mxint8.tar.gz -C /your/target/folder

8. Download the code demo.py in this repository

Download or directly copy the code demo.py in this repository

9. Demo

python demo.py

10. Example: