From fbc3a1907aeb6beff59461e535045f17ac14306e Mon Sep 17 00:00:00 2001 From: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Date: Fri, 12 Jun 2026 09:38:12 -0400 Subject: [PATCH] [Bug] Migrate Reset cache for both v2 and v1 model runner (#42759) Signed-off-by: yewentao256 Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> --- vllm/v1/worker/gpu/model_runner.py | 2 -- vllm/v1/worker/gpu_model_runner.py | 3 +++ 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/vllm/v1/worker/gpu/model_runner.py b/vllm/v1/worker/gpu/model_runner.py index d269bf25bdb..328b521bfc8 100644 --- a/vllm/v1/worker/gpu/model_runner.py +++ b/vllm/v1/worker/gpu/model_runner.py @@ -367,8 +367,6 @@ class GPUModelRunner(LoRAModelRunnerMixin): from vllm.v1.worker.gpu_model_runner import GPUModelRunner as GPUModelRunnerV1 GPUModelRunnerV1.reload_weights(self, *args, **kwargs) # type: ignore[arg-type] - self.reset_encoder_cache() - self.reset_mm_cache() def apply_sparse_weight_patches(self, *args, **kwargs) -> None: # TODO: Use full version instead of import when fully migrated to v2 diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py index f3f52c75d8b..cb607c0b7b0 100644 --- a/vllm/v1/worker/gpu_model_runner.py +++ b/vllm/v1/worker/gpu_model_runner.py @@ -5391,6 +5391,9 @@ class GPUModelRunner( weights_not_loaded, ) + self.reset_encoder_cache() + self.reset_mm_cache() + def _get_prompt_logprobs_dict( self, hidden_states: torch.Tensor,