TetraJet-v2-NVFP4Training
Python
★ 13
updated 1mo ago
[ICML 2026 Spotlight] Official implementation of TetraJet-v2: Accurate NVFP4 Training for LLMs, with fully-NVFP4 linear layer with unbiased backprop, and algorithms to overcome LLMs' weight-oscillation and activation-outlier bottlenecks.
No plain-English explanation yet — one is being written right now. Check back in a minute.