LLM

• Observability
• Vector Databases
• OWASP Top 10 - Doc - Slides
• Unstructured - ETL for LLMs
• LoRA = low rank adaptation - vastly cheaper mechanism for fine tuning
• Emerging Architectures for LLM Applications
• https://github.com/imartinez/privateGPT
• Lit-GPT
• Jina AI
• https://github.com/Stability-AI/StableLM - Stability AI #OpenSource
• https://github.com/Torantulino/Auto-GPT
• https://github.com/hpcaitech/ColossalAI
  • https://www.hpc-ai.tech/blog/one-half-day-of-training-using-a-few-hundred-dollars-yields-similar-results-to-mainstream-large-models-open-source-and-commercial-free-domain-specific-llm-solution
• Transformer for Actions - GPT - https://www.adept.ai/blog/act-1
• Toolformer
  • Meta desenvolve modelo de linguagem que aprende sozinho a usar ferramentas externas, como mecanismos de busca, calculadoras e calendários
  • Decide por conta própria quais APIs usar para contornar as limitações presentes em outros sistemas, como a falta de conhecimento em aritmética, por exemplo. Seu desempenho é superior ao GPT-3, mesmo sendo baseado no GPT-J, um modelo com apenas 4% dos parâmetros.
  • https://arstechnica.com/information-technology/2023/02/meta-develops-an-ai-language-bot-that-can-use-external-software-tools/
  • https://arxiv.org/abs/2302.04761
• You probably don’t know how to do Prompt Engineering
• https://github.com/Lightning-AI/lit-llama #OpenSource

Good speech-to-text models have this trait, along with language translation programs and on-screen swipe keyboards. In each of these cases we want to be understood, not surprised. AI, therefore, makes the most sense as a translation layer between humans, who are incurably chaotic, and traditional software, which is deterministic.

an adaptive interface between chaotic real-world problems and secure, well-architected technical solutions. AI may not truly understand us, but it can deliver our intentions to an API with reasonable accuracy and describe the results in a way we understand.

from https://stackoverflow.blog/2023/05/01/ai-isnt-the-app-its-the-ui/

ChatGPT Link to heading

• Plugins
  • https://medium.com/geekculture/6-chatgpt-mind-blowing-extensions-to-use-it-anywhere-db6638640ec7
    • https://github.com/wong2/chat-gpt-google-extension
    • https://github.com/gragland/chatgpt-chrome-extension
  • https://platform.openai.com/docs/plugins/introduction - Docs
  • http://openai.com/waitlist/plugins - Waitlist
  • https://openai.com/blog/chatgpt-plugins#OpenAI - Landing Page
  • https://github.com/openai/chatgpt-retrieval-plugin - The open-source retrieval plugin enables ChatGPT to access personal or organizational information sources

Code generation Link to heading

• https://github.com/replit/ReplitLM
• https://huggingface.co/blog/starcoder
  • https://huggingface.co/bigcode/starcoder
  • https://marketplace.visualstudio.com/items?itemName=HuggingFace.huggingface-vscode

Copilot Link to heading

- https://marketplace.visualstudio.com/items?itemName=gencay.vscode-chatgpt - ChatGPT no VSCode

• Copilot Internals - Reverse engineering

Comparison Link to heading

• Artificial Analysis - independent analysis of AI models and hosting providers
• Chatbot Arena - Benchmarking LLMs in the Wild

Datasets Link to heading

• https://pile.eleuther.ai/ - An 800GB Dataset of Diverse Text for Language Modeling
• https://laion.ai/ - Large-scale Artificial Intelligence Open Network - TRULY OPEN AI. 100% NON-PROFIT. 100% FREE.

Hosting Link to heading

Cloud Link to heading

• https://www.cerebrium.ai/ - makes it easier to train, deploy and monitor machine learning models with just a few lines of code - Serverless GPU Model Deployment
• https://cohere.ai/ - build high performance, secure LLM for the enterprise - powerful capabilities, like content generation, summarization, and search
• CoreWeave - cloud provider, delivering a massive scale of GPUs.
• Foundry - Instant Compute ML infra.
• Lambda - access to GPUs for deep learning.
• Modal - Run generative AI models, large-scale batch jobs, job queues, and much more.
• Monster API - access powerful generative AI models with our auto-scaling APIs, zero management required.
• Replicate - Run models in the cloud at scale.

Decentralized Link to heading

• GPUtopia - GPU Marketplace
• Petals - Run large language models at home, BitTorrent‑style - Repo

Local Link to heading

• LocalAI - Self-hosted, community-driven, local OpenAI-compatible API. (…) No GPU required…
• llama.cpp
  • Python bindings - Repo
  • A comprehensive guide to running Llama 2 locally
• MLC LLM - LLM on iPhone - App Store - Repository
• MLX - An array framework for Apple silicon

Image Generation Link to heading

• Clipdrop
• DALL-E 2
  • Prompt book
  • DALL·E 2 Prompt Engineering Guide
  • Use DALL-E to create infinite zoom movies
• DreamStudio
• Hidden in Plain Sight - Hide text into images
• Midjourney
• Stable Diffusion
  • Demo Hugging Face
  • Online
  • Outpainting on an infinite canvas - Tweet and Sourcecode
• Runway
• https://stockimg.ai/
• https://www.tryleap.ai/

Models Link to heading

• OpenHermes 2.5 - Mistral 7B - can do function calling #OpenSource

MPT Link to heading

• Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs
• https://github.com/mosaicml/composer - PyTorch library that enables you to train neural networks faster, at lower cost, and to higher accuracy - Python
• https://github.com/mosaicml/streaming - make training on large datasets from cloud storage as fast, cheap, and scalable as possible. Python
• https://github.com/mosaicml/llm-foundry

Low-code platforms Link to heading

• Rivet - The Open-Source Visual AI Programming Environment #OpenSource
• Stack AI - The No-Code AI Automation Platform
• Vellum - The dev platform for production LLM apps

RAG Link to heading

• ColBERT - fast and accurate retrieval model, enabling scalable BERT-based search over large text collections in tens of milliseconds.
  • Exploring ColBERT with RAGatouille
  • RAGatouille - bridging the gap between state-of-the-art research and alchemical RAG pipeline practices

SDKs Link to heading

• Guardrails - lets a user add structure, type and quality guarantees to the outputs of LLMs.
• Guidance - Python lib that allows you to interleave generation, prompting, and logical control into a single continuous flow matching #OpenSource
• Kor - Python lib that “helps” you extract structured data from text using LLMs #OpenSource
• LangChain - Python lib to develop AI applications - PromptTemplate, LLMs interface, etc. #OpenSource
  • Agents use an LLM to determine which actions to take and in what order.
  • Pros
    • Builtin Chain serialization
  • BabyAGI #Pinecone
  • Langflow - UI designed with react-flow to provide an effortless way to experiment and prototype flows.
• LiteLLM - Call 100+ LLMs using the same Input/Output format #OpenSource
• LlamaIndex - framework for RAG #OpenSource
• LLM - A CLI utility and Python lib for interacting with OpenAI, PaLM and local models installed on your own machine #OpenSource
• MonkeyPatch - easily call an LLM in place of the function body in Python. The more you use MonkeyPatch functions, the cheaper and faster they gets (up to 9-10x!) through automatic model distillation. #OpenSource
• Prompt Engine - Javascript lib for creating and maintaining prompts #OpenSource
• Semantic Kernel - Python and C# libs that allow to define plugins that can be chained together #OpenSource

Speech Recognition Link to heading

• https://github.com/openai/whisper - general-purpose speech recognition model
  • https://github.com/ggerganov/whisper.cpp - runs on the CPU
    • https://huggingface.co/ggerganov/whisper.cpp/tree/main - models