From 48f18788dc437af2a89fa062c5bde06cef05fc78 Mon Sep 17 00:00:00 2001
From: Alexander Dallmann <alexander.dallmann@uni-wuerzburg.de>
Date: Mon, 30 Sep 2024 09:13:53 +0200
Subject: [PATCH] Fix: ScaledDotProductAttention.forward throws AttributeError
 when executed within `enable_torch_sdp`

---
 curated_transformers/layers/attention.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/curated_transformers/layers/attention.py b/curated_transformers/layers/attention.py
index d72c9f60..0ea0875a 100644
--- a/curated_transformers/layers/attention.py
+++ b/curated_transformers/layers/attention.py
@@ -708,7 +708,7 @@ def forward(
                 key=key,
                 value=value,
                 attn_mask=logit_mask,
-                dropout_p=self.dropout_prob if self.training else 0.0,
+                dropout_p=self.dropout.p if self.training else 0.0,
             )
 
             # Torch SDP returns NaNs for pieces where every is piece masked out.