Pinned Loading

1. Kilat Kilat Public

   Lightweight transformer toolkit, hybrid attention (global decay + latent MLA), MoE FFN, KV-cache inference, and production training loop. Built for LLM research.

   Python 1

2. suara suara Public

   Suara is the alternative of attention by utilizing FFT as the mixer, nevertheless it still perform poorly this repo shown an initial research of causal wave convolution

   Python 1

3. SuaraBERT SuaraBERT Public

   Again This is only the experiments of Causal Wave Convolution

   Python

4. gawa gawa Public

   GAWA (Gaussian-Weighted Abstraction for Words) is a character-level morphological encoder–decoder designed to represent words as dense, fixed-size vectors without relying on traditional tokenizatio…

   Jupyter Notebook