A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Core] Drop overly aggressive whisper assertion (#25408)
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Russell Bryant committed
Sep 22, 2025 at 20:09 UTC
f31ff874607c311dda7aebe52f73c20f5cad923d
1 file changed
-4
vllm/v1/core/sched/scheduler.py
-4
@@ -463,10 +463,6 @@ class Scheduler(SchedulerInterface):
463
# always padded to the maximum length. If we support other
464
# encoder-decoder models, this will need to be updated if we
465
# want to only allocate what is needed.
466
- assert ("whisper"
467
- in self.vllm_config.model_config.model.lower()), (
468
- "Whisper is the only supported "
469
- "encoder-decoder model.")
466
num_encoder_tokens =\
467
self.scheduler_config.max_num_encoder_input_tokens
468
else: