From 1e989784d37903ae449e4000bb5085927fbf282a Mon Sep 17 00:00:00 2001
From: hailin <hailin@gdzx.xyz>
Date: Thu, 28 Aug 2025 22:54:46 +0800
Subject: [PATCH] .

---
 train_sft_ds.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/train_sft_ds.py b/train_sft_ds.py
index c255748..e6bd4cb 100644
--- a/train_sft_ds.py
+++ b/train_sft_ds.py
@@ -487,7 +487,7 @@ def main():
         torch_dtype=dtype,
         low_cpu_mem_usage=True,
         trust_remote_code=True,
-        attn_implementation="flash_attention_2"
+        attn_implementation="sdpa"
     )
 
     print(f"GC enabled? {getattr(model, 'is_gradient_checkpointing', False)}", flush=True)