[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
+9 -13
View File
@@ -64,25 +64,21 @@ class NeuronWorker(LocalOrDistributedWorkerBase):
assert (self.lora_config
is None), ("LoRA is not supported for TransformersNeuronX "
"framework.")
from vllm.worker.multi_step_neuron_model_runner import (
MultiStepNeuronModelRunner)
if self.speculative_config is not None:
return MultiStepNeuronModelRunner(vllm_config=vllm_config)
else:
return NeuronModelRunner(vllm_config=vllm_config)
raise NotImplementedError(
"Speculative decoding is not supported for TransformersNeuronX"
)
return NeuronModelRunner(vllm_config=vllm_config)
def get_neuronx_distributed_model_runner(self, vllm_config):
from vllm.worker.multi_step_neuronx_distributed_model_runner import (
MultiStepNeuronxDistributedModelRunner)
from vllm.worker.neuronx_distributed_model_runner import (
NeuronxDistributedModelRunner)
if self.speculative_config is not None:
assert (self.lora_config
is None), "LoRA is not supported for Speculative Decoding"
return MultiStepNeuronxDistributedModelRunner(
vllm_config=vllm_config)
else:
return NeuronxDistributedModelRunner(vllm_config=vllm_config)
assert (self.lora_config is None), (
"LoRA is not supported for Speculative Decoding")
raise NotImplementedError(
"Speculative decoding is not supported for NeuronxDistributed")
return NeuronxDistributedModelRunner(vllm_config=vllm_config)
def init_device(self) -> None:
self.init_distributed_environment()