From 3116e028d28e08a48b21d667af8dd051ecf09c4b Mon Sep 17 00:00:00 2001
From: nihui <nihuini@tencent.com>
Date: Fri, 27 Oct 2023 11:50:52 +0800
Subject: [PATCH] pnnx drop sdap scale=None for compatiblity with old torch
 (#5107)

---
 .../pass_level2/F_scaled_dot_product_attention.cpp    | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/tools/pnnx/src/pass_level2/F_scaled_dot_product_attention.cpp b/tools/pnnx/src/pass_level2/F_scaled_dot_product_attention.cpp
index 8dcfafaf12b..ecfc6d8acaf 100644
--- a/tools/pnnx/src/pass_level2/F_scaled_dot_product_attention.cpp
+++ b/tools/pnnx/src/pass_level2/F_scaled_dot_product_attention.cpp
@@ -65,6 +65,17 @@ pnnx.Output             output      1 0 out
     {
         return "F.scaled_dot_product_attention";
     }
+
+    void write(Operator* op, const std::map<std::string, Parameter>& captured_params, const std::map<std::string, Attribute>& captured_attrs) const
+    {
+        GraphRewriterPass::write(op, captured_params, captured_attrs);
+
+        if (captured_params.at("scale").type == 0)
+        {
+            // drop scale=None for compatiblity with old torch
+            op->params.erase("scale");
+        }
+    }
 };
 
 REGISTER_GLOBAL_PNNX_GRAPH_REWRITER_PASS(F_scaled_dot_product_attention_1, 10)