From 10fa9eea21ae757d17c1369afa6172598db3be92 Mon Sep 17 00:00:00 2001 From: youkaichao Date: Thu, 16 May 2024 11:07:41 -0700 Subject: [PATCH] [Misc] remove old comments (#4866) --- vllm/worker/model_runner.py | 10 ---------- 1 file changed, 10 deletions(-) diff --git a/vllm/worker/model_runner.py b/vllm/worker/model_runner.py index 460f98d7..cd7af256 100644 --- a/vllm/worker/model_runner.py +++ b/vllm/worker/model_runner.py @@ -887,16 +887,6 @@ class ModelRunner: # This usually takes < 10 seconds. logger.info("Graph capturing finished in %.0f secs.", elapsed_time) - def __del__(self) -> None: - # Delete the CUDA graphs before deleting the pynccl communicator. - # NOTE(woosuk): This is necessary because otherwise deadlocks can - # happen. - # FIXME(woosuk): This is a bit hacky. Find a more robust solution. - # TODO(youkaichao): when we get enough user feedback that pynccl is - # more stable than cupy, we can remove this, e.g. in v0.4.1. - self.graph_runners.clear() - self.pynccl_backend = None - @property def vocab_size(self) -> int: return self.model_config.get_vocab_size()