Match in-place vLLM MoE serving path
This commit is contained in:
@@ -171,7 +171,7 @@ def main() -> None:
|
|||||||
w2=w2,
|
w2=w2,
|
||||||
topk_weights=topk_weights,
|
topk_weights=topk_weights,
|
||||||
topk_ids=topk_ids,
|
topk_ids=topk_ids,
|
||||||
inplace=False,
|
inplace=True,
|
||||||
use_fp8_w8a8=True,
|
use_fp8_w8a8=True,
|
||||||
per_channel_quant=False,
|
per_channel_quant=False,
|
||||||
global_num_experts=global_num_experts,
|
global_num_experts=global_num_experts,
|
||||||
|
|||||||
Reference in New Issue
Block a user