AI-Hypercomputer / AI-Hypercomputer/maxtext

bug: Numeric divergence between the mHC layer and its Pallas kernel

Aperta
#5,195 2 commenti 0 reazioni 0 assegnatari Vedi su GitHub
bug
Lingua principale
Python
Stelle
2.4k
Fork
607
Merge medio
2g 19h
PR unite (30g)
158

Descrizione

### Bug report

DeepSeek v4. The reference path had its own mapping helper that cast alpha_scale and
beta down to self.dtype before computing sigmoid(alpha_scale * h + beta),
while the Pallas path called common.compute_sigmoid_gate, which keeps the
scale and bias in float32. With dtype: bfloat16 the two branches therefore
gated on different numbers, so toggling use_mhc_pallas_kernel changed the
layer's output beyond accumulation noise.
Fixed in PR:
https://github.com/AI-Hypercomputer/maxtext/pull/5179

### Logs/Output

_No response_

### Environment Information

_No response_

### Additional Context

_No response_

Guida per i contributori

Apri la guida per i contributori

Valutazione

Questa issue non è ancora stata valutata.

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.