EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Improve code styles (#4021)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-03-03 03:20:23 -08:00
committed by GitHub
parent ac2387279e
commit 66301e124f
14 changed files with 88 additions and 243 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
benchmark/kernels/fused_moe_triton/benchmark_torch_compile_fused_moe.py
View File

@@ -30,6 +30,11 @@ def get_model_config(model_name: str, tp_size: int):
topk = config.num_experts_per_tok
intermediate_size = config.moe_intermediate_size
shard_intermediate_size = 2 * intermediate_size // tp_size
elif config.architectures[0] in ["DeepseekV2ForCausalLM", "DeepseekV3ForCausalLM"]:
E = config.n_routed_experts
topk = config.num_experts_per_tok
intermediate_size = config.intermediate_size
shard_intermediate_size = 2 * intermediate_size // tp_size
elif config.architectures[0] in [
"Grok1ForCausalLM",
"Grok1ImgGen",
@@ -39,11 +44,6 @@ def get_model_config(model_name: str, tp_size: int):
topk = config.num_experts_per_tok
intermediate_size = config.moe_intermediate_size
shard_intermediate_size = 2 * intermediate_size // tp_size
elif config.architectures[0] in ["DeepseekV2ForCausalLM", "DeepseekV3ForCausalLM"]:
E = config.n_routed_experts
topk = config.num_experts_per_tok
intermediate_size = config.intermediate_size
shard_intermediate_size = 2 * intermediate_size // tp_size
else:
# Default: Mixtral
E = config.num_local_experts