Seeed Studio 🔥Software Backbone of SenseCAP Watcher, Next Physical AI Agent for Smarter Space

Finetune Llama3-8B model with NVIDIA Robotics Jetson AGX Orin with specific Harry Potter Fanatic dataset created using the GPT-4o model! Project on Hackster👉 https://www.hackster.io/shahizat/ai-assistant-for-harry-potter-fans-using-nvidia-jetson-board-4c6fbc

Shakhizat Nurgaliyev showcased the whole pipeline of model fine-tuning and inferencing deployment steps, where you can enter multiple prompts to get conversational responses continuously:
✅ define training arguments for fine-tuning Llama3;
✅ further model adjustments for better performance;
✅ utilize the MLC LLM(Machine Learning Compiler Large Language Model) method to run inferencing.

Curious how to leverage MLC LLM to accelerate the inferencing speed of the model? Check out our wiki to quantized Llama2 on Jetson Orin: https://wiki.seeedstudio.com/Quantized_Llama2_7B_with_MLC_LLM_on_Jetson/

✨ This project integrates AI and RGB lighting to create a smart, interactive display. Using components such as an LED matrix, Grove Vision AI Module V2, XIAO ESP32C3 (or even ESP32), and a Raspberry Pi Camera.

The project uses WLED for LED control, SenseCraft for image processing, and a custom script for gesture recognition. One standout feature is its 100% local processing, ensuring privacy and data security. Alejandro's detailed guide walks you through the process, from setting up the hardware to configuring the software and uploading the hand detection model to the vision module. 📸