CompactifAI/Full-Chunked-KL-Loss

Fully chunked KL-Loss kernel, that massively reduces VRAM usage

2 best points · 1 comments · 1 Hacker News post · 20 GitHub stars · View on GitHub

Hacker News discussions

• Chunked KL loss, running Knowledge Distillation locally in less <6GB VRAM (2 points, 1 comments, by ikergarcia1996, 2026-08-11)