From 6c15945e4dadfe8aaddf03396aeb9474bc77d309 Mon Sep 17 00:00:00 2001 From: xiaoxiaohehe001 <49090790+xiaoxiaohehe001@users.noreply.github.com> Date: Wed, 15 Oct 2025 16:19:39 +0800 Subject: [PATCH] fix_fa3 (#4429) --- fastdeploy/model_executor/layers/attention/flash_attn_backend.py | 1 + 1 file changed, 1 insertion(+) diff --git a/fastdeploy/model_executor/layers/attention/flash_attn_backend.py b/fastdeploy/model_executor/layers/attention/flash_attn_backend.py index df5e5db9d..e9871d6f7 100644 --- a/fastdeploy/model_executor/layers/attention/flash_attn_backend.py +++ b/fastdeploy/model_executor/layers/attention/flash_attn_backend.py @@ -125,6 +125,7 @@ class FlashAttentionBackend(AttentionBackend): self.pd_disaggregation_mode: str = fd_config.parallel_config.pd_disaggregation_mode self.start_layer_index: int = fd_config.model_config.start_layer_index + self.rope_3d: bool = getattr(fd_config.model_config, "rope_3d", False) if fd_config.parallel_config.expert_parallel_rank is None: fd_config.parallel_config.expert_parallel_rank = 0