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
[Misc] Remove unused slot_mapping buffer (#23502)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Woosuk Kwon committed
Aug 24, 2025 at 14:03 UTC
ad78868450a3596bed37dac05be9049019953e94
1 file changed
-3
vllm/v1/worker/gpu_model_runner.py
-3
@@ -254,9 +254,6 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
254
self.seq_lens = torch.zeros(self.max_num_reqs,
255
dtype=torch.int32,
256
device=self.device)
257
- self.slot_mapping = torch.zeros(self.max_num_tokens,
258
- dtype=torch.int64,
259
- device=self.device)
257
258
# None in the first PP rank. The rest are set after load_model.
259
self.intermediate_tensors: Optional[IntermediateTensors] = None