[V0 Deprecation] Remove multi-step scheduling (#22138)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Woosuk Kwon <woosuk@thinkingmachines.ai>
This commit is contained in:
Woosuk Kwon
2025-08-12 20:18:39 -07:00
committed by GitHub
parent e18859298d
commit 71683ca6f6
37 changed files with 57 additions and 3465 deletions
+2 -12
View File
@@ -118,20 +118,10 @@ class CudaPlatformBase(Platform):
@classmethod
def check_and_update_config(cls, vllm_config: "VllmConfig") -> None:
parallel_config = vllm_config.parallel_config
scheduler_config = vllm_config.scheduler_config
model_config = vllm_config.model_config
if parallel_config.worker_cls == "auto":
if scheduler_config.is_multi_step:
if envs.VLLM_USE_V1:
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 vllm_config.speculative_config:
if not envs.VLLM_USE_V1:
raise NotImplementedError(
"Speculative decoding is not supported on vLLM V0.")
@@ -139,7 +129,7 @@ class CudaPlatformBase(Platform):
else:
if envs.VLLM_USE_V1:
parallel_config.worker_cls = \
"vllm.v1.worker.gpu_worker.Worker"
"vllm.v1.worker.gpu_worker.Worker"
else:
parallel_config.worker_cls = "vllm.worker.worker.Worker"
+2 -12
View File
@@ -327,18 +327,8 @@ class RocmPlatform(Platform):
cache_config.block_size = 16
parallel_config = vllm_config.parallel_config
scheduler_config = vllm_config.scheduler_config
if parallel_config.worker_cls == "auto":
if scheduler_config.is_multi_step:
if envs.VLLM_USE_V1:
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 vllm_config.speculative_config:
if not envs.VLLM_USE_V1:
raise NotImplementedError(
"Speculative decoding is not supported on vLLM V0.")
@@ -346,7 +336,7 @@ class RocmPlatform(Platform):
else:
if envs.VLLM_USE_V1:
parallel_config.worker_cls = \
"vllm.v1.worker.gpu_worker.Worker"
"vllm.v1.worker.gpu_worker.Worker"
else:
parallel_config.worker_cls = "vllm.worker.worker.Worker"
+1 -6
View File
@@ -133,18 +133,13 @@ class TpuPlatform(Platform):
parallel_config = vllm_config.parallel_config
scheduler_config = vllm_config.scheduler_config
if parallel_config.worker_cls == "auto":
if scheduler_config.is_multi_step:
raise NotImplementedError(
"Multi-step scheduling is not supported (and not "
"needed) on vLLM V1. Please launch without "
"--num-scheduler-steps.")
parallel_config.worker_cls = "vllm.v1.worker.tpu_worker.TPUWorker"
assert not vllm_config.speculative_config, (
"Speculative decoding is not yet supported for TPU backend")
if scheduler_config.is_multimodal_model and not \
scheduler_config.disable_chunked_mm_input:
scheduler_config.disable_chunked_mm_input:
logger.warning("TPU does not support running Multimodal models"\
" without setting `--disable_chunked_mm_input`. " \
"Forcing --disable_chunked_mm_input.")