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, 22:35 UTC
[Model] Remove DSV2 unused code (#21903)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Jul 30, 2025 at 15:55 UTC
fc91da549978347a3b5f5ebe6e8cbeae6148e012
1 file changed
-14
vllm/model_executor/models/deepseek_v2.py
-14
@@ -830,20 +830,6 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP, MixtureOfExperts):
830
sampling_metadata)
831
return logits
832
833
- def make_empty_intermediate_tensors(
834
- self, batch_size: int, dtype: torch.dtype,
835
- device: torch.device) -> IntermediateTensors:
836
- return IntermediateTensors({
837
- "hidden_states":
838
- torch.zeros((batch_size, self.config.hidden_size),
839
- dtype=dtype,
840
- device=device),
841
- "residual":
842
- torch.zeros((batch_size, self.config.hidden_size),
843
- dtype=dtype,
844
- device=device),
845
- })
846
-
833
def load_weights(self, weights: Iterable[tuple[str,
834
torch.Tensor]]) -> set[str]:
835
stacked_params_mapping = [