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

Add typo checker in pre-commit (#6179)

Browse Source 
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
This commit is contained in:
applesaucethebun
2025-05-11 00:55:00 -04:00
committed by GitHub
parent de167cf5fa
commit 2ce8793519
99 changed files with 154 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/models/deepseek_v2.py
View File

@@ -1287,7 +1287,7 @@ class DeepseekV2DecoderLayer(nn.Module):
# Fully Connected
hidden_states = self.mlp(hidden_states)
# TODO(ch-wan): ues reduce-scatter in MLP to avoid this scatter
# TODO(ch-wan): use reduce-scatter in MLP to avoid this scatter
# Scatter
if self.dp_size != 1:
# important: forward batch.gathered_buffer is used both after scatter and after gather.
@@ -1499,7 +1499,7 @@ class DeepseekV2ForCausalLM(nn.Module):
else:
assert (
self.n_share_experts_fusion == self.tp_size
), f"Shared experts fusion optimization is enabled in DeepSeek V3/R1, set it to {self.tp_size} can get best optimized performace."
), f"Shared experts fusion optimization is enabled in DeepSeek V3/R1, set it to {self.tp_size} can get best optimized performance."
elif self.n_share_experts_fusion == 0:
if (
_is_cuda
@@ -1665,7 +1665,7 @@ class DeepseekV2ForCausalLM(nn.Module):
if is_nextn:
if hasattr(self.config, "num_nextn_predict_layers"):
num_nextn_layers = self.config.num_nextn_predict_layers
assert num_nextn_layers == 1, "Only 1 nextn layer is supportted"
assert num_nextn_layers == 1, "Only 1 nextn layer is supported"
# compatible with old design
nextn_layer_id = (
0