ZIBEE/bi_150-vllm
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Upgrade to vllm 0.17.0 corex v4.1 overlay

Browse Source
This commit is contained in:
LZiBee
2026-04-29 19:38:22 +08:00
parent 8fac6062e4
commit 938d0854a5
430 changed files with 35969 additions and 14511 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/compilation/passes/fusion/collective_fusion.py
View File

@@ -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,