A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bench] Add qwen-next in benchmark_moe.py (#24661)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Sep 11, 2025 at 21:29 UTC
d11ec124a0409a78a23539212356510c79c2111e
1 file changed
+5
-1
benchmarks/kernels/benchmark_moe.py
+5
-1
@@ -594,7 +594,11 @@ def main(args: argparse.Namespace):
594
E = config.n_routed_experts
595
topk = config.num_experts_per_tok
596
intermediate_size = config.moe_intermediate_size
597
- elif config.architectures[0] in ("Qwen2MoeForCausalLM", "Qwen3MoeForCausalLM"):
597
+ elif config.architectures[0] in (
598
+ "Qwen2MoeForCausalLM",
599
+ "Qwen3MoeForCausalLM",
600
+ "Qwen3NextForCausalLM",
601
+ ):
602
E = config.num_experts
603
topk = config.num_experts_per_tok
604
intermediate_size = config.moe_intermediate_size