[Core] Drop overly aggressive whisper assertion (vllm-project#25408)

Signed-off-by: Russell Bryant <[email protected]>
Signed-off-by: xuebwang-amd <[email protected]>

Author: russellb

vllm/v1/core/sched/scheduler.py

@@ -463,10 +463,6 @@ def schedule(self) -> SchedulerOutput:
                     # always padded to the maximum length. If we support other
                     # encoder-decoder models, this will need to be updated if we
                     # want to only allocate what is needed.
-                    assert ("whisper"
-                            in self.vllm_config.model_config.model.lower()), (
-                                "Whisper is the only supported "
-                                "encoder-decoder model.")
                     num_encoder_tokens =\
                         self.scheduler_config.max_num_encoder_input_tokens
                 else: