From d5c0f43b86c221d40cc8eba12cc79b24c8168642 Mon Sep 17 00:00:00 2001 From: Roger Wang Date: Wed, 24 Sep 2025 02:47:30 -0700 Subject: [PATCH] [Bugfix] Fix dummy video number of frames calculation (#25553) Signed-off-by: Roger Wang Signed-off-by: yewentao256 --- vllm/v1/worker/gpu_model_runner.py | 2 +- vllm/v1/worker/tpu_model_runner.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py index eebdbcc621c69..ee339e22cea90 100644 --- a/vllm/v1/worker/gpu_model_runner.py +++ b/vllm/v1/worker/gpu_model_runner.py @@ -2807,7 +2807,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin): dummy_decoder_data = self.mm_registry.get_decoder_dummy_data( model_config=self.model_config, - seq_len=self.max_num_tokens, + seq_len=self.max_model_len, mm_counts={modality: 1}, cache=self.mm_budget.cache, ) diff --git a/vllm/v1/worker/tpu_model_runner.py b/vllm/v1/worker/tpu_model_runner.py index 4cbf991a14c11..4a2adb1e6510d 100644 --- a/vllm/v1/worker/tpu_model_runner.py +++ b/vllm/v1/worker/tpu_model_runner.py @@ -1795,7 +1795,7 @@ class TPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin): dummy_decoder_data = self.mm_registry.get_decoder_dummy_data( model_config=self.model_config, - seq_len=self.max_num_tokens, + seq_len=self.max_model_len, mm_counts={modality: 1}, cache=self.mm_budget.cache, )