LLM training in simple, raw C/CUDA

Tile primitives for speedy kernels

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

[ICML2025] SpargeAttention: A training-free sparse attention that accelerates any model inference.

CPM.cu is a lightweight, high-performance CUDA implementation for LLMs, optimized for end-device inference and featuring cutting-edge techniques in sparse architecture, speculative sampling and qua…