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
[Model] fix DeepSeek e_score_correction_bias dtype to fp32 (#23640)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Aug 26, 2025 at 20:09 UTC
fdeb3dac132c9ef92d981dd811529e6496781b07
1 file changed
+1
-1
vllm/model_executor/models/deepseek_v2.py
+1
-1
@@ -126,7 +126,7 @@ class DeepseekV2MoE(nn.Module):
126
prefix=f"{prefix}.gate")
127
if config.topk_method == "noaux_tc":
128
self.gate.e_score_correction_bias = nn.Parameter(
129
- torch.empty(config.n_routed_experts))
129
+ torch.empty(config.n_routed_experts, dtype=torch.float32))
130
else:
131
self.gate.e_score_correction_bias = None
132