TrickyRex
/

bitserial-modmul-v8

modular-arithmetic

neural-arithmetic

Model card Files Files and versions

TrickyRex commited on Jun 28

Commit

b49812c

·

verified ·

1 Parent(s): 63aa9f0

Update README.md

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -12,7 +12,7 @@ entry_class model.BitSerialReducer, output_base 2, ~471K params.
 - Deterministic (re-run identical); bf16 decisions match fp32 exactly (0 flips, min|logit|=3.0).
 - Randomizing the weights collapses every tier to 0.00 (capability is in the trained parameters).
-## Limitation (honest)
 Benchmark-accurate on the official random-operand distribution; a sparse, power-of-two-adjacent
 structured-input residual remains (not exact). Not organizer-verified. The Lean package proves the integer algorithm,
 not this network.

 - Deterministic (re-run identical); bf16 decisions match fp32 exactly (0 flips, min|logit|=3.0).
 - Randomizing the weights collapses every tier to 0.00 (capability is in the trained parameters).
+## Limitations
 Benchmark-accurate on the official random-operand distribution; a sparse, power-of-two-adjacent
 structured-input residual remains (not exact). Not organizer-verified. The Lean package proves the integer algorithm,
 not this network.