Pinned Loading

1. research research Public

   LM Research

   Python

2. GraphZero GraphZero Public

   Graph Zero is a novel AlphaZero-style algorithm incorporating Graph Attention Networks for improved game state representation, implemented in pure CUDA, distributed training

   Cuda 1

3. NanoTitan NanoTitan Public

   A project focused on maximizing HellaSwag scores on models trained on a budget (also a pretraining codebase). How high can a 124M parameter base model trained on 10B tokens for 1 epoch get? Which a…

   Python 1

4. cuLLM cuLLM Public

   Large Language Model and Tokenizer written from scratch in CUDA, LLaMA3.1 architecture. Tokenizer written from scratch in C++, GPT-4o tokenizer.

   Cuda 1

5. BitDB BitDB Public

   SQLite Database Clone written in one C file. Added LoRA fine-tuned Mistral-7B model written in MLX for natural language to database query support.

   C 2