[diffusion] UX: deduplicate SM120 FP8 fallback warnings across layers (#41542)
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com>
This commit is contained in:
@@ -68,7 +68,6 @@ class FP8FlashAttentionSM120Impl(AttentionImpl):
|
||||
prefix=prefix,
|
||||
**extra_impl_args,
|
||||
)
|
||||
self._reported_fallbacks: set[str] = set()
|
||||
|
||||
# --- dispatch -------------------------------------------------------------
|
||||
|
||||
@@ -89,10 +88,7 @@ class FP8FlashAttentionSM120Impl(AttentionImpl):
|
||||
return None
|
||||
|
||||
def _report_fallback(self, reason: str) -> None:
|
||||
if reason in self._reported_fallbacks:
|
||||
return
|
||||
self._reported_fallbacks.add(reason)
|
||||
logger.warning(
|
||||
logger.warning_once(
|
||||
"fp8_fa_sm120 attention: %s; using cuDNN SDPA for these calls.", reason
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user