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 15, 14:45 UTC
[Misc] Remove redundant comment (#16703)
Signed-off-by: Jade Zheng <zheng.shoujian@outlook.com>
Jade Zheng committed
Apr 17, 2025 at 08:44 UTC
8a7368e0692fd1e355e85e196b5d3509343ebabf
1 file changed
-3
vllm/v1/worker/gpu_model_runner.py
-3
@@ -540,9 +540,6 @@ class GPUModelRunner(LoRAModelRunnerMixin):
540
# because M (max_model_len) is not necessarily divisible by block_size.
541
block_table_indices = (req_indices * self.max_num_blocks_per_req +
542
positions_np // self.block_size)
543
- # NOTE(woosuk): We use torch.index_select instead of np.take here
544
- # because torch.index_select is much faster than np.take for large
545
- # tensors.
543
block_table_cpu = self.input_batch.block_table.get_cpu_tensor()
544
block_numbers = block_table_cpu.flatten()[block_table_indices].numpy()
545
block_offsets = positions_np % self.block_size