(trivial) remove redundant

yuanheng-zhao · yuanheng-zhao · commit 0ec90af9de05 · 2024-01-18T17:44:46.000+08:00
diff --git a/colossalai/kernel/triton/flash_decoding.py b/colossalai/kernel/triton/flash_decoding.py
@@ -269,8 +269,7 @@ def flash_decoding_attention(
         HEAD_DIM=head_dim,
     )
 
-    output = torch.empty_like(q)  # already overlapped
-    output = torch.empty((bsz, 1, num_heads, head_dim), dtype=q.dtype, device=q.device)
+    output = torch.empty((bsz, 1, num_heads, head_dim), dtype=q.dtype, device=q.device)  # already overlapped
 
     grid = (bsz, num_heads)
     _flash_decoding_fwd_reduce_kernel[grid](

Original file line number	Diff line number	Diff line change
`@@ -269,8 +269,7 @@ def flash_decoding_attention(`
`269`	`269`	`HEAD_DIM=head_dim,`
`270`	`270`	`)`
`271`	`271`
`272`		`- output = torch.empty_like(q) # already overlapped`
`273`		`- output = torch.empty((bsz, 1, num_heads, head_dim), dtype=q.dtype, device=q.device)`
	`272`	`+ output = torch.empty((bsz, 1, num_heads, head_dim), dtype=q.dtype, device=q.device) # already overlapped`
`274`	`273`
`275`	`274`	`grid = (bsz, num_heads)`
`276`	`275`	`_flash_decoding_fwd_reduce_kernel[grid](`