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

Clean logs for DeepSeek-V3 launching (#6079)

Browse Source
This commit is contained in:
Baizhou Zhang
2025-05-07 18:54:50 -07:00
committed by GitHub
parent 8f508cc77f
commit 73600673bb
7 changed files with 61 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
python/sglang/srt/models/deepseek_v2.py
View File

@@ -88,6 +88,7 @@ from sglang.srt.utils import (
get_int_env_var,
is_cuda,
is_hip,
log_info_on_rank0,
)
_is_hip = is_hip()
@@ -1485,8 +1486,9 @@ class DeepseekV2ForCausalLM(nn.Module):
):
self.n_share_experts_fusion = 0
global_server_args_dict["n_share_experts_fusion"] = 0
logger.info(
"Only Deepseek V3/R1 can use shared experts fusion optimization. Shared experts fusion optimization is disabled."
log_info_on_rank0(
logger,
"Only Deepseek V3/R1 can use shared experts fusion optimization. Shared experts fusion optimization is disabled.",
)
else:
assert (
@@ -1501,8 +1503,9 @@ class DeepseekV2ForCausalLM(nn.Module):
):
self.n_share_experts_fusion = self.tp_size
global_server_args_dict["n_share_experts_fusion"] = self.tp_size
logger.info(
"Deepseek V3/R1 with fp8 can use shared experts fusion optimization when SM version >=90. Shared experts fusion optimization is enabled."
log_info_on_rank0(
logger,
"Deepseek V3/R1 with fp8 can use shared experts fusion optimization when SM version >=90. Shared experts fusion optimization is enabled.",
)
def get_input_embeddings(self) -> nn.Embedding: