Pinned Loading

1. distillreasoning distillreasoning Public

   Teaching a 4B model to think by borrowing reasoning from models 200x its size.

   Python

2. tinyllm tinyllm Public

   I built a tiny LLM from scratch to understand how GPT-4 and LLaMA actually work. 10M params, trained on Shakespeare, modernized with RMSNorm + SwiGLU + RoPE + KV cache. Every mistake documented.

   Python

3. recallforge recallforge Public

   Cross-modal vision-language search engine. Index text and images, search across modalities. Multi-backend (PyTorch/MLX), tiered modes, MCP server. Inspired by QMD by Tobi.

   Python 1

4. mollygraph mollygraph Public

   Embedded local-first graph and vector memory for AI agents, with Ladybug storage, GLiNER2 extraction, and MCP/API/SDK access.

   Python 1