From 0f71ec8656374b80da728c55217a2a073fac9ef5 Mon Sep 17 00:00:00 2001 From: Mick Date: Fri, 2 Oct 2026 19:25:15 +0800 Subject: [PATCH] [diffusion] UX: deduplicate SM120 FP8 fallback warnings across layers (#41542) Co-authored-by: Mick Qian --- .../runtime/layers/attention/backends/fp8_fa_sm120_attn.py | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/python/sglang/multimodal_gen/runtime/layers/attention/backends/fp8_fa_sm120_attn.py b/python/sglang/multimodal_gen/runtime/layers/attention/backends/fp8_fa_sm120_attn.py index b8faa26f75..126a253522 100644 --- a/python/sglang/multimodal_gen/runtime/layers/attention/backends/fp8_fa_sm120_attn.py +++ b/python/sglang/multimodal_gen/runtime/layers/attention/backends/fp8_fa_sm120_attn.py @@ -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 )