Pinned Loading

1. PocketGPT PocketGPT Public

   SOTA GPT-style language model built from scratch with PyTorch Lightning, featuring a custom Muon-AdamW optimizer, self trained BPE tokenizer, GQA, alternate value embeddings, XSA. Pretrained on 5B …

   Python 3

2. Triton-MuonAdamW Triton-MuonAdamW Public

   Hybrid Muon + AdamW optimizer with custom fused Triton kernels for fast, drop-in training.

   Python

3. FREUID-Challenge-2026---IJCAI-ECAI FREUID-Challenge-2026---IJCAI-ECAI Public

   ConvNeXt-based document fraud detection model for the FREUID Challenge 2026 (IJCAI-ECAI).

   HTML

4. GPT-Poem GPT-Poem Public

   A GPT-style decoder-only Transformer built and trained from scratch for poetry generation. The model is implemented in PyTorch without using the built-in attention modules—multi-head self-attention…

   Jupyter Notebook