NLP

Build a Large Language Model From Scratch

Hands-on book and code repository teaching LLM internals from the ground up. Covers pretraining, fine-tuning, and RLHF with clear PyTorch implementations.

by Sebastian Raschka

Pinecone AI Education & Canopy

Developer education content and open-source RAG framework for Pinecone. Tutorials on vector search, embeddings, and building production RAG systems.

by James Briggs

sentence-transformers

The most popular library for computing dense vector representations of text. Powers semantic search, clustering, and similarity across millions of applications.

by Nils Reimers

TRL (Transformer Reinforcement Learning)

Hugging Face library for training language models with RLHF, DPO, and PPO. The standard open-source toolkit for alignment fine-tuning.

by Lewis Tunstall

Hugging Face Inference Endpoints

Production deployment platform for transformer models. One-click deployment of any Hugging Face model with auto-scaling and optimized inference.

by Philipp Schmid

GPT-NeoX / Pythia

EleutherAI's open-source large language model suite. Pythia provides 16 model checkpoints (70M-12B) for scientific study of LLM training dynamics.

by Stella Biderman

Together AI Platform

Cloud platform for training, fine-tuning, and running open-source AI models. Offers the fastest inference for Llama, Mixtral, and other open models.

by Vinicius Moens

Mistral AI Models

Open-weight frontier models from Europe. Mistral 7B, Mixtral 8x7B, and Mistral Large compete with models 10x their size through architecture innovations.

by Arthur Mensch

Sakana AI — Evolutionary Model Merging

Nature-inspired AI research lab using evolutionary algorithms to automatically merge and create new foundation models without training from scratch.

by David Ha

Imbue AI Agents

AI systems that reason about code and build software autonomously. Imbue trains models optimized for practical reasoning and agent capabilities.

by Kanjun Qiu

Adept ACT-1

AI model trained to use software tools like a human. ACT-1 can navigate web browsers, use spreadsheets, and operate enterprise software via natural language.

by David Luan

Jasper AI

AI content generation platform for enterprise marketing teams. Used by 100K+ businesses to create on-brand copy, blog posts, and campaigns at scale.

by Dave Rogenmoser

Character.AI

Platform for creating and chatting with AI characters. Over 20M monthly users interact with millions of user-created AI personalities and assistants.

by Daniel De Freitas

Inflection Pi / Microsoft AI

Personal AI assistant designed for emotional intelligence and natural conversation. Pi pioneered empathetic AI before Mustafa moved to lead Microsoft AI.

by Mustafa Suleyman

Perplexity AI

AI-native answer engine with real-time web search and citations. Redefining search by providing direct, sourced answers instead of link lists.

by Aravind Srinivas

AssemblyAI Universal-2

State-of-the-art speech-to-text API with speaker diarization, sentiment analysis, and summarization. Universal-2 model achieves best-in-class accuracy.

by Dylan Fox

Deepgram Nova-2

Enterprise speech AI platform with real-time transcription, custom vocabulary, and language understanding. Nova-2 delivers industry-leading speed and accuracy.

by Scott Stephenson

Cohere

Enterprise LLM platform built by a Transformer paper co-author. RAG, embeddings, and language understanding for enterprise search and generation.

by Aidan Gomez

Allen AI (AI2)

Nonprofit AI research lab founded by Paul Allen. Builds open-source AI tools: Semantic Scholar, OLMo, AI2 THOR. Mission: AI for the common good.

by Ali Farhadi

Semantic Scholar

AI-powered academic search engine by Allen AI. Indexes 200M+ papers with AI-generated TLDRs, citation contexts, and research recommendations.

by Michael Schmitz

Lexion AI

AI contract management for enterprises. Uses NLP to extract key terms, track obligations, and automate legal document workflows.

by Ammad Ahmad

OLMo

Fully open language model from Allen AI. Open training data (Dolma), open code, open weights, open evals. Making LLM science reproducible.

by Dirk Groeneveld

Textio

Augmented writing platform. NLP predicts how language performs in job posts, emails, and business writing. Used by Fortune 500 for inclusive hiring.

by Suchen Zang

Megatron-LM

NVIDIA's framework for training large transformer language models efficiently across thousands of GPUs using model and data parallelism.

by Bryan Catanzaro

PyTorch

The most widely used open-source deep learning framework. Provides dynamic computation graphs, GPU acceleration, and a rich ecosystem for research and production.

by Soumith Chintala

Hugging Face Transformers

The definitive open-source library for state-of-the-art NLP, vision, and multimodal models. Provides 200K+ pretrained models and unified APIs.

by Thomas Wolf

RAG (Retrieval-Augmented Generation)

The original RAG framework combining retrieval and generation for knowledge-intensive NLP. Now a foundational pattern used across the LLM ecosystem.

by Patrick Lewis

LoRA

Low-Rank Adaptation of Large Language Models. The most widely adopted parameter-efficient fine-tuning method, reducing trainable parameters by 10,000x.

by Edward Hu

bitsandbytes / QLoRA

Efficient 4-bit quantization library enabling fine-tuning of 65B parameter models on a single 48GB GPU. QLoRA made LLM fine-tuning accessible to everyone.

by Tim Dettmers

FlashAttention

IO-aware exact attention algorithm that is 2-4x faster and uses 5-20x less memory. Now integrated into PyTorch and used by virtually all LLM training runs.

by Tri Dao

Mamba

Selective state space model architecture offering linear-time sequence modeling. Achieves transformer-quality performance with 5x faster inference throughput.

by Albert Gu

Qwen

Alibaba's family of large language and multimodal models. Qwen2 series achieves top performance among open-weight models across benchmarks.

by Junyang Lin

Reka AI Models

Multimodal AI models from Reka AI. Reka Core, Flash, and Edge deliver frontier performance across text, image, video, and audio understanding.

by Yi Tay

PaLM Training Infrastructure

Scaling infrastructure behind Google's PaLM model. Achieved efficient training of 540B parameter models across 6144 TPU v4 chips.

by Reiner Pope

Open Assistant

Community-driven open-source project to create a free, high-quality chat assistant dataset and model. One of the earliest open RLHF efforts.

by Christoph Schuhmann

Lil'Log

Comprehensive technical blog covering LLMs, diffusion, RL, and more. The most widely cited personal ML blog, serving as a reference for researchers worldwide.

by Lilian Weng

Contextual AI RAG 2.0

Enterprise RAG platform that goes beyond naive retrieval. Contextual AI builds RAG-native language models trained end-to-end for grounded generation.

by Douwe Kiela

DSPy

Programming framework for optimizing LLM pipelines. Replaces hand-written prompts with composable, learnable modules that auto-optimize via compilation.

by Omar Khattab

ALiBi Positional Encoding

Attention with Linear Biases enables transformers to generalize to longer sequences than seen during training without learned position embeddings.

by Ofir Press

Machine Learning Mastery

The largest practitioner-focused ML education platform with 1000+ tutorials and 20+ books covering deep learning, NLP, computer vision, and time series.

by Jason Brownlee

OpenAI API Platform

Built and scaled OpenAI's API platform serving GPT-4, DALL-E, and Whisper to millions of developers. Architected the product powering ChatGPT.

by Peter Welinder

OpenAI Engineering

Co-founded OpenAI and led engineering from GPT-1 through ChatGPT and GPT-4. Built the infrastructure and team behind the most impactful AI products.

by Greg Brockman

Thinking Machines Lab

AI research startup founded after departing OpenAI as CTO. Led the technical development of ChatGPT, GPT-4, and DALL-E at OpenAI.

by Mira Murati

Scale AI Data Platform

Enterprise data labeling and AI infrastructure platform. Powers training data for OpenAI, Meta, and US DoD with human-in-the-loop annotation at scale.

by Alexandr Wang

Hugging Face Transformers

The most popular open-source library for state-of-the-art NLP, vision, and audio models. Hosts 500K+ models and 100K+ datasets. The GitHub of machine learning.

by Clem Delangue

AI Grants & Investments

Portfolio of AI investments and open-source contributions. Co-leads Pioneer Fund. Previously acquired GitHub for Microsoft and shipped Copilot.

by Nat Friedman

OpenAI Cookbook & DevRel

Built OpenAI's developer ecosystem from scratch. Created the OpenAI Cookbook with 200+ examples, grew the API community to millions of developers.

by Logan Kilpatrick

Latent Space & smol.ai

Latent Space is the leading AI engineering podcast and newsletter. smol.ai builds small, useful AI developer tools. Defined the AI Engineer role.

by Swyx (Shawn Wang)

fast.ai

Free deep learning courses and the fastai library. Made cutting-edge deep learning accessible to anyone who can code. Used by hundreds of thousands of students.

by Jeremy Howard

ML Paper Explanations (YouTube)

YouTube channel with 250K+ subscribers explaining cutting-edge ML papers. Co-created OpenAssistant, an open-source ChatGPT alternative.

by Yannic Kilcher

Replit

AI-powered coding platform used by 30M+ developers. Features Replit Agent for building full-stack apps from natural language prompts in the browser.

by Amjad Masad

Chain-of-Thought Research

Pioneering research on chain-of-thought prompting, instruction tuning, and emergent abilities of large language models at OpenAI.

by Jason Wei

v0 by Vercel

AI-powered UI generation. Describe a component in natural language, get production-ready React/Tailwind code. Uses LLMs for code generation.

by Guillermo Rauch

llm

CLI tool for interacting with Large Language Models. Supports OpenAI, Claude, local models via plugins. Log and search prompts.

by Simon Willison

DocETL

AI-powered data processing pipelines for unstructured documents. LLM-driven extraction, transformation, and analysis at scale.

by Shreya Shankar

nanoGPT

The simplest, fastest repository for training/finetuning medium-sized GPTs. Reproduces GPT-2 (124M) in ~$20 of compute.

by Andrej Karpathy

MosaicML / DBRX

Open-source efficient LLM training platform. DBRX is a 132B MoE model rivaling GPT-3.5. Acquired by Databricks for $1.3B.

by Jonathan Frankle

LangChain

Framework for developing applications powered by language models. Connects LLMs to external data, tools, and agents.

by Harrison Chase