wassnameandClaudypoo f22bbeb0d2 parity(U1): cached-J pullback vs direct VJP, all layers cos>0.999 ...

min cos 0.999801 (layer 8), rising to 0.999996; gap is fp16 cache
storage as expected. Gate PASS.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

2026-07-10 12:52:18 +08:00

..

parity_u1.py

parity(U1): cached-J pullback vs direct VJP, all layers cos>0.999

2026-07-10 12:52:18 +08:00

smoke.py

smoke: fit+steer Qwen3-0.6B on happy/joy end-to-end

2026-07-10 12:51:08 +08:00