Open Source LLMOps

Experience Open Source Large Language Models (LLMs)

• Master cutting-edge LLM architectures like Transformers through hands-on labs
• Fine-tune models on your data with SkyPilot's scalable training platform
• Deploy efficiently with model servers like LoRAX and vLLM

Explore the Open Source LLM Ecosystem:

• Gain in-depth understanding of how LLMs work under the hood
• Run pre-trained models like Code Llama, Mistral & Stable Diffusion
• Discover advanced architectures like Sparse Expert Models
• Launch cloud GPU instances for accelerated compute

Guided LLM Project:

• Fine-tune LLaMA, Mistral or other LLMs on your custom dataset
• Leverage SkyPilot to scale training across cloud providers
• Containerize your fine-tuned model for production deployment
• Serve models efficiently with LoRAX, vLLM and other open servers
• Build powerful AI solutions leveraging state-of-the-art open source language models. Gain practical LLMOps skills through code-first learning.

Week 1: Getting Started with Open Source Ecosystem

• Introduction to popular open source natural language processing models and their capabilities

• Accessing pre-trained NLP models using libraries like HuggingFace Transformers

• Using large language models for synthetic data augmentation to enhance datasets

• Building real-world NLP solutions using open source tools in Python and Rust

Week 2: Using Local LLMs from LLamaFile to Whisper.cpp

• Key components of LLamaFile for packaging language models into portable files

• Running local language models from LLamaFile on your own devices

• Automating speech recognition workflows using Whisper.cpp

• Integrating Whisper.cpp into GenAI building blocks and applications

Week 3: Applied Projects

• Using language models in the browser with Transformers.js and ONNX

• Exporting models to the ONNX format for enhanced portability

• Developing portable command-line interfaces with the Cosmopolitan project

• Building a phrase generator application as a native binary using Cosmopolitan

Week 4: Recap and Final Challenges

• Connecting to local language models with APIs using Python

• Retrieval augmented generation using local LLMs

• Hands-on labs for GPU-accelerated MLOps workflows

• Final project to build an interactive LLamaFile sandbox

By the end of this course, learners will have gained practical experience leveraging state-of-the-art open source language models to build AI applications. They will be able to deploy solutions on their own devices as well as integrate models into efficient MLOps pipelines.