Deep dives into AI research — from paper breakdowns to new ideas, with code and math.
A deep-dive into Kimi's Attention Residuals paper — why standard residual connections dilute layer contributions — and a novel extension to LoRA fine-tuning: AttnRes-LoRA, with math and code.
Read postStay tuned for upcoming tutorials on diffusion models, 3D Gaussian Splatting, RLHF, and more.