From ffce72c0415701eb10af86668cf504a5eeba2a72 Mon Sep 17 00:00:00 2001 From: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Date: Wed, 10 Jun 2026 14:06:01 -0400 Subject: [PATCH] [Model Runner V2] Fix v2 `AttributeError: 'CohereASRDecoder' object has no attribute 'embed_input_ids'` (#44568) Signed-off-by: yewentao256 Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> --- vllm/v1/worker/gpu/model_states/__init__.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/vllm/v1/worker/gpu/model_states/__init__.py b/vllm/v1/worker/gpu/model_states/__init__.py index 06b5a92c395..b096fcaf5e6 100644 --- a/vllm/v1/worker/gpu/model_states/__init__.py +++ b/vllm/v1/worker/gpu/model_states/__init__.py @@ -13,7 +13,10 @@ def init_model_state( encoder_cache: EncoderCache | None, device: torch.device, ): - if "WhisperForConditionalGeneration" in vllm_config.model_config.architectures: + if ( + "WhisperForConditionalGeneration" in vllm_config.model_config.architectures + or "CohereAsrForConditionalGeneration" in vllm_config.model_config.architectures + ): from vllm.v1.worker.gpu.model_states.whisper import WhisperModelState return WhisperModelState(vllm_config, model, encoder_cache, device)