[V0 Deprecation] Remove async_output_proc, preemption mode, delay factor (#25334)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
@@ -126,10 +126,6 @@ class CpuPlatform(Platform):
|
||||
"""
|
||||
torch.cpu.set_device(device)
|
||||
|
||||
@classmethod
|
||||
def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
|
||||
return False
|
||||
|
||||
@classmethod
|
||||
def inference_mode(cls):
|
||||
return torch.no_grad()
|
||||
|
||||
@@ -96,16 +96,6 @@ class CudaPlatformBase(Platform):
|
||||
def get_device_total_memory(cls, device_id: int = 0) -> int:
|
||||
raise NotImplementedError
|
||||
|
||||
@classmethod
|
||||
def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
|
||||
if enforce_eager and not envs.VLLM_USE_V1:
|
||||
logger.warning(
|
||||
"To see benefits of async output processing, enable CUDA "
|
||||
"graph. Since, enforce-eager is enabled, async output "
|
||||
"processor cannot be used")
|
||||
return False
|
||||
return True
|
||||
|
||||
@classmethod
|
||||
def is_fully_connected(cls, device_ids: list[int]) -> bool:
|
||||
raise NotImplementedError
|
||||
|
||||
@@ -275,13 +275,6 @@ class Platform:
|
||||
"""Get the total memory of a device in bytes."""
|
||||
raise NotImplementedError
|
||||
|
||||
@classmethod
|
||||
def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
|
||||
"""
|
||||
Check if the current platform supports async output.
|
||||
"""
|
||||
raise NotImplementedError
|
||||
|
||||
@classmethod
|
||||
def inference_mode(cls):
|
||||
"""A device-specific wrapper of `torch.inference_mode`.
|
||||
|
||||
@@ -310,16 +310,6 @@ class RocmPlatform(Platform):
|
||||
device_props = torch.cuda.get_device_properties(device_id)
|
||||
return device_props.total_memory
|
||||
|
||||
@classmethod
|
||||
def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
|
||||
if enforce_eager and not envs.VLLM_USE_V1:
|
||||
logger.warning(
|
||||
"To see benefits of async output processing, enable CUDA "
|
||||
"graph. Since, enforce-eager is enabled, async output "
|
||||
"processor cannot be used")
|
||||
return False
|
||||
return True
|
||||
|
||||
@classmethod
|
||||
def check_and_update_config(cls, vllm_config: "VllmConfig") -> None:
|
||||
from vllm.config.compilation import CUDAGraphMode
|
||||
|
||||
@@ -75,10 +75,6 @@ class TpuPlatform(Platform):
|
||||
def get_device_total_memory(cls, device_id: int = 0) -> int:
|
||||
raise NotImplementedError
|
||||
|
||||
@classmethod
|
||||
def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
|
||||
return False
|
||||
|
||||
@classmethod
|
||||
def get_punica_wrapper(cls) -> str:
|
||||
return "vllm.lora.punica_wrapper.punica_tpu.PunicaWrapperTPU"
|
||||
|
||||
@@ -98,10 +98,6 @@ class XPUPlatform(Platform):
|
||||
device_props = torch.xpu.get_device_properties(device_id)
|
||||
return device_props.total_memory
|
||||
|
||||
@classmethod
|
||||
def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
|
||||
return True
|
||||
|
||||
@classmethod
|
||||
def inference_mode(cls):
|
||||
return torch.no_grad()
|
||||
|
||||
Reference in New Issue
Block a user