minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

fix: make override DeepseekV2Model work (#14707)

Browse Source
This commit is contained in:
Yineng Zhang
2025-12-08 23:11:18 -08:00
committed by GitHub
parent 0f8bd55f3e
commit b626334475
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/models/deepseek_v2.py
+1 -2
View File
@@ -3202,7 +3202,6 @@ class DeepseekV2Model(nn.Module):
class DeepseekV2ForCausalLM(nn.Module): class DeepseekV2ForCausalLM(nn.Module):
# for quark model load # for quark model load
packed_modules_mapping = {} packed_modules_mapping = {}
model_cls = DeepseekV2Model
def __init__( def __init__(
self, self,
@@ -3228,7 +3227,7 @@ class DeepseekV2ForCausalLM(nn.Module):
self.quant_config = quant_config self.quant_config = quant_config
self.determine_num_fused_shared_experts() self.determine_num_fused_shared_experts()
self.use_nsa = is_deepseek_nsa(config) self.use_nsa = is_deepseek_nsa(config)
self.model = self.model_cls( self.model = DeepseekV2Model(
config, quant_config, prefix=add_prefix("model", prefix) config, quant_config, prefix=add_prefix("model", prefix)
) )
if self.pp_group.is_last_rank: if self.pp_group.is_last_rank: