From 5f671cb4c3145194e94ffb393ee459432f7fa2b8 Mon Sep 17 00:00:00 2001 From: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> Date: Tue, 28 Jan 2025 23:56:56 -0500 Subject: [PATCH] [V1] Improve Error Message for Unsupported Config (#12535) Co-authored-by: Michael Goin --- vllm/platforms/cuda.py | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/vllm/platforms/cuda.py b/vllm/platforms/cuda.py index 2587e3a11dde3..e4b436edf7588 100644 --- a/vllm/platforms/cuda.py +++ b/vllm/platforms/cuda.py @@ -120,13 +120,18 @@ class CudaPlatformBase(Platform): if parallel_config.worker_cls == "auto": if scheduler_config.is_multi_step: if envs.VLLM_USE_V1: - raise NotImplementedError + raise NotImplementedError( + "Multi-step scheduling is not supported (and not " + "needed) on VLLM V1. Please launch without " + "--num-scheduler-steps.") else: parallel_config.worker_cls = \ "vllm.worker.multi_step_worker.MultiStepWorker" elif vllm_config.speculative_config: if envs.VLLM_USE_V1: - raise NotImplementedError + raise NotImplementedError( + "Speculative decoding is not yet supported on VLLM V1." + ) else: parallel_config.worker_cls = \ "vllm.spec_decode.spec_decode_worker.create_spec_worker"