MaximilianKr/llm-internship

Material, resources, overview of LLMs

Material, resources, overview of LLMs

• General
  • Tools
• LangChain
  • DeepLearning.AI courses
  • Example Implementations / Guides
    • Python
    • Javascript
• Llama 2
  • German
• Embeddings
• Vector Database
• Deployment
• Finetuning
• Other Material
  • Prompt Injection
  • Educational

General

• Open LLMs for Commercial Use
• Huggingface Open LLM Leaderboard
• Build ChatGPT from GPT-3
• Comparison of self-hosted LLMs and OpenAI: cost, text generation quality, development speed, privacy

Tools

• text-generation-webui (fast inference on most local models)
• LocalAI (OpenAI drop-in alternative REST API)
• BionicGPT (onprem ChatGPT replacement)
• OpenAccess-AI-Collective/axolotl (fine-tune local models)
• LLaMA-Factory (easy finetuning framework for popular local models)
• Petals (bittorrent-like deployment of local models)

LangChain

• ChatLangChain (interactive documentation)
• LangSmith (debugging, evaluating, monitoring agents)
• LangServe (deploy LangChain code)
• OpenGPTs (OSS alternative to OpenAIs Custom GPTs)

DLAI courses

• LangChain - LangChain for LLM Application Development
• LangChain - Chat with Your Data

The notebooks include several examples on how to use LangChain and are located in DeepLearning.AI Notebooks. Follow the instructions in the README.md in there to set up your environment and use either your OpenAI API key or LocalAI to run the notebooks.

Example Implementations / Guides

Python

• LangChain Local LLMs
• LangChain + Streamlit + Llama Guide
• Llama v2.0, FAISS in Python using LangChain
• Answering Question About your Documents Using LangChain
• Using LLaMA 2.0, FAISS and LangChain for Question-Answering
• Unlocking the Power of LangChain: Building Applications with Large Language Models

Without LangChain:

• Building RAG from Scratch (with LlamaIndex)

Javascript

• Building an AI chatbot with Next.js, Langchain, and OpenAI
• LangChain Deployment with Next.js

Llama 2

• How to prompt Llama 2
• llama.cpp (also bindings for Python)
• Talk-Llama (local implementation with llama.cpp and whisper)
• Quantize Llama models with GGML and llama.cpp
• TinyLlama (still being trained)

German

• Other Languages Support
  • TLDR: Llama2 usage in other than English out-of-scope
  • solutions:
    • append "Answer in German" to initial prompt (in English)
    • prompt in German
    • use fine-tuned model for German (beware of licensing and quality)

Fine-tuned Models

• SauerkrautLM
• LeoLM
• German Assistant
• EM German LEO Mistral (quantized also available)

Embeddings

• Getting started with embeddings
• SBERT Transformer Embedding Models (local)
• HF text-embeddings-inference (local container)
• Chat with Anything - From PDFs Files to Image Documents
  • Notebook with examples on file type detection, chunking, embedding

Vector Database

Hosted

• Pinecone (free tier available for one index/project)

Local

• ChromaDB
• Weaviate

Deployment

• LangChain Guide Deployment
• LangChain Deployment Templates
• Frameworks for Serving LLMs

Finetuning

• RAG vs Finetuning
• DLAI Course Finetuning
• Finetuning Digital Twins
• LangChain Finetune model in your voice (from chat data)
• Fine-Tune Your Own Llama 2 Model in a Colab Notebook

Other Material

Prompt Injection

• What’s the worst that can happen?
• LLM Attacks

Educational

• FCC Agile Software Development Handbook
• Vicki Boykis: What Are Embeddings?
• Transformers Code step by step
• The Illustrated Transformer
• Stanford NLU Spring 2023 Playlist
• reddit/Starter Guide for Playing with Your Own Local AI