Commit 55f7ee1
committed
R21v1: Deploy PR openai#1536 (VarLen + Doc-Independent LoRA TTT, no Triton)
PR openai#1536 base code unmodified. 2756 lines, 112KB.
VarLen Flash Attention + Doc-Independent LoRA TTT.
No Triton, no CUTLASS. Banking with PyTorch fallback MLP.
Muon 0.97 already default. Expected: ~1.0775 BPP.1 parent 9058206 commit 55f7ee1
1 file changed
+2756
-2
lines changed
0 commit comments