Grouped Query Experts: Mixture-of-Experts on GQA Self-Attention Paper • 2606.20945 • Published 7 days ago • 60
FrontiersMind/Nandi-Mini-V1.1-600M-Intermediate-Checkpoint-400GT Text Generation • 0.6B • Updated 26 days ago • 323 • 8
view article Article How I contributed a new model to the Transformers library using Codex nielsr • Mar 30 • 52
Running on CPU Upgrade 262 The Synthetic Data Playbook: Generating Trillions of the Finest Tokens 📝 262 Visualize synthetic‑data experiments as an interactive bookshelf
view article Article The 1 Billion Token Challenge: Finding the Perfect Pre-training Mix codelion • Nov 3, 2025 • 65
Running Featured 1.37k FineWeb: decanting the web for the finest text data at scale 🍷 1.37k Explore and download the FineWeb web‑scale text dataset
Running 3.9k The Ultra-Scale Playbook 🌌 3.9k The ultimate guide to training LLM on large GPU Clusters
The Instruction Gap: LLMs get lost in Following Instruction Paper • 2601.03269 • Published Dec 19, 2025 • 8
Running on CPU Upgrade Featured 3.22k The Smol Training Playbook 📚 3.22k The secrets to building world-class LLMs
view reply You don't really have to clone the repo. The FastAPI code is just there for demonstration, and you can code the way you like. The main takeaway is the Dockerfile.