The Smol Training Playbook
📚
3.21k
The secrets to building world-class LLMs
🧠 Brain Cache
A curated collection of my favorite AI and ML research papers — one place to find the papers I keep coming back to.
-
-
FineVision: Open Data is All You Need
📝227A new open-source dataset for training VLMs
-
physics-intern: an Autonomous Agent for Physics Research
📝52Explore an autonomous AI workflow for physics research
-
The Synthetic Data Playbook: Generating Trillions of the Finest Tokens
📝259Visualize synthetic‑data experiments as an interactive bookshelf
-
Token-In, Token-Out Done Right
🧩6Explore an interactive simulation while reading the article
-
Evaluation Guidebook
📝329Explore LLM benchmark scores over time
-
The Ultra-Scale Playbook
🌌3.9kThe ultimate guide to training LLM on large GPU Clusters
-
Maintain the unmaintainable
📚81Explore the complex relationships between 400+ machine learning models
-
FineWeb: decanting the web for the finest text data at scale
🍷1.37kExplore and download the FineWeb web‑scale text dataset
-
Porting nanochat to Transformers: an AI modeling history lesson
📝49Learn about ML and Transformers through nanochat
-
FinePDFs: Liberating 3T of the finest tokens from PDFs
📄75 -
Can LLMs Play the Game of Science?
📝28Explore LLM science benchmark scores
-
QED-Nano: Teaching a Tiny Model to Prove Hard Theorems
📝74Who needs 1T parameters? Olympiad proofs with a 4B model
-
Distilling 100B+ Models 40x Faster with TRL
📝88TRL distillation for 100B+ teachers, 40x faster
-
Extracting Signal from the Noise
📝12Explore PR categories with an interactive particle visualization
-
The ultimate guide to RL environments: building and scaling them in the LLM era
📝188Building and scaling RL environments for LLM training
-
Defeating the trainer-generator precision mismatch in TRL
🎯19Download research PDF (Pro access required)