regolith
strata but optimized for m5 (and maybe more later)
Star
Fork
Watch

Round 3 blitz: GDN decode/prefill kernel optimization (gdn_step -12%, gdn_scanr_4 -8.4%), r1 MoCE unscaling fix, spec_min_p schedule (32.3-40.3 tok/s decode, 325 tok/s prefill)

Author dawn Date Commit Parent 4b0f233f
+730 -108
4 changed files
Unified Split
+730 -108
4 changed files