Unverified Commit 07bdabef authored by Wang Xingran's avatar Wang Xingran Committed by GitHub
Browse files

[Bugfix] Use 'sum' reduction instead of 'avg' in Async TP reduce-scatter (#33088)


Signed-off-by: default avatarXingran Wang <wangxingran123456@outlook.com>
Signed-off-by: default avatarHongjian Zhang <hirokenovo@gmail.com>
Co-authored-by: default avatarHongjian Zhang <hirokenovo@gmail.com>
parent a572baff
......@@ -53,7 +53,7 @@ class GEMMReduceScatterPattern(BasePattern):
gemm_rs = torch.ops.symm_mem.fused_matmul_reduce_scatter(
mul,
mm_weight,
"avg",
"sum",
scatter_dim=0,
group_name=self.tp.device_group.group_name,
)
......@@ -150,7 +150,7 @@ class ScaledMMReduceScatterPattern(BasePattern):
mat2,
scale_a,
scale_b,
"avg",
"sum",
scatter_dim, # orig_scatter_dim
scatter_dim, # scatter_dim_after_maybe_reshape
self.tp.device_group.group_name,
......@@ -285,7 +285,7 @@ class CutlassScaledMMReduceScatterPattern(BasePattern):
mat2,
scale_a,
scale_b,
"avg",
"sum",
scatter_dim, # orig_scatter_dim
scatter_dim, # scatter_dim_after_maybe_reshape
self.tp.device_group.group_name,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment