[museformer] fix parameter problem

museformer/museformer/attention/__init__.py

@@ -28,7 +28,7 @@ def create_attention_v2_s1(
         return BlocksparseRpeSelfAttentionV2S1(
             *args, block_size=block_size, **kwargs
         )
-    return NotImplementedError(implementation)
+    raise NotImplementedError(implementation)
 
 
 def create_attention(

museformer/museformer/museformer_lm.py

@@ -119,7 +119,7 @@ def base_lm_architecture(args):
         args.sum2con = ((None,),)
         args.sum2sum = ((None,),)
 
-    args.attention_impl = getattr(args, "attention_impl", "triton")
+    args.attention_impl = getattr(args, "attention_impl", "blocksparse")
     if args.attention_impl == 'blocksparse':
         args.block_size = getattr(args, "block_size", 64)