biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[VLM] Refactor MultiModalConfig initialization and profiling (#7530)

Browse Source
This commit is contained in:
Roger Wang
2024-08-17 13:30:55 -07:00
committed by GitHub
parent 1ef13cf92f
commit bbf55c4805
29 changed files with 143 additions and 190 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
vllm/executor/cpu_executor.py
View File

@@ -141,7 +141,6 @@ class CPUExecutor(ExecutorBase):
rank=rank,
distributed_init_method=self.distributed_init_method,
lora_config=self.lora_config,
multimodal_config=self.multimodal_config,
kv_cache_dtype=self.cache_config.cache_dtype,
prompt_adapter_config=self.prompt_adapter_config,
is_driver_worker=rank == 0,

6
vllm/executor/executor_base.py
View File

@@ -2,8 +2,8 @@ from abc import ABC, abstractmethod
from typing import List, Optional, Set, Tuple
from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
ModelConfig, MultiModalConfig, ObservabilityConfig,
ParallelConfig, PromptAdapterConfig, SchedulerConfig,
ModelConfig, ObservabilityConfig, ParallelConfig,
PromptAdapterConfig, SchedulerConfig,
SpeculativeConfig)
from vllm.lora.request import LoRARequest
from vllm.prompt_adapter.request import PromptAdapterRequest
@@ -29,7 +29,6 @@ class ExecutorBase(ABC):
device_config: DeviceConfig,
load_config: LoadConfig,
lora_config: Optional[LoRAConfig],
multimodal_config: Optional[MultiModalConfig],
speculative_config: Optional[SpeculativeConfig],
prompt_adapter_config: Optional[PromptAdapterConfig],
observability_config: Optional[ObservabilityConfig],
@@ -41,7 +40,6 @@ class ExecutorBase(ABC):
self.parallel_config = parallel_config
self.scheduler_config = scheduler_config
self.device_config = device_config
self.multimodal_config = multimodal_config
self.speculative_config = speculative_config
self.prompt_adapter_config = prompt_adapter_config
self.observability_config = observability_config

1
vllm/executor/gpu_executor.py
View File

@@ -55,7 +55,6 @@ class GPUExecutor(ExecutorBase):
rank=rank,
distributed_init_method=distributed_init_method,
lora_config=self.lora_config,
multimodal_config=self.multimodal_config,
speculative_config=self.speculative_config,
prompt_adapter_config=self.prompt_adapter_config,
is_driver_worker=(not self.parallel_config)

1
vllm/executor/openvino_executor.py
View File

@@ -49,7 +49,6 @@ class OpenVINOExecutor(ExecutorBase):
rank=0,
distributed_init_method=distributed_init_method,
lora_config=self.lora_config,
multimodal_config=self.multimodal_config,
kv_cache_dtype=self.cache_config.cache_dtype,
is_driver_worker=True,
)

8
vllm/executor/ray_xpu_executor.py
View File

@@ -7,9 +7,8 @@ from typing import (TYPE_CHECKING, Any, Awaitable, Dict, List, Optional, Set,
import vllm.envs as envs
from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
ModelConfig, MultiModalConfig, ParallelConfig,
PromptAdapterConfig, SchedulerConfig,
SpeculativeConfig)
ModelConfig, ParallelConfig, PromptAdapterConfig,
SchedulerConfig, SpeculativeConfig)
from vllm.executor.distributed_gpu_executor import ( # yapf: disable
DistributedGPUExecutor, DistributedGPUExecutorAsync)
from vllm.executor.ray_utils import RayWorkerWrapper, ray
@@ -46,7 +45,6 @@ class RayXPUExecutor(DistributedGPUExecutor):
device_config: DeviceConfig,
load_config: LoadConfig,
lora_config: Optional[LoRAConfig],
multimodal_config: Optional[MultiModalConfig],
prompt_adapter_config: Optional[PromptAdapterConfig],
speculative_config: Optional[SpeculativeConfig],
) -> None:
@@ -61,7 +59,6 @@ class RayXPUExecutor(DistributedGPUExecutor):
self.parallel_config = parallel_config
self.scheduler_config = scheduler_config
self.device_config = device_config
self.multimodal_config = multimodal_config
self.prompt_adapter_config = prompt_adapter_config
placement_group = self.parallel_config.placement_group
@@ -203,7 +200,6 @@ class RayXPUExecutor(DistributedGPUExecutor):
rank=rank,
distributed_init_method=distributed_init_method,
lora_config=self.lora_config,
multimodal_config=self.multimodal_config,
is_driver_worker=rank == 0,
))
self._run_workers("init_worker", all_kwargs=init_worker_all_kwargs)

1
vllm/executor/tpu_executor.py
View File

@@ -52,7 +52,6 @@ class TPUExecutor(ExecutorBase):
local_rank=local_rank,
rank=rank,
distributed_init_method=distributed_init_method,
multimodal_config=self.multimodal_config,
is_driver_worker=rank == 0,
)

7
vllm/executor/xpu_executor.py
View File

@@ -3,9 +3,8 @@ from typing import List, Optional
import torch
from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
ModelConfig, MultiModalConfig, ParallelConfig,
PromptAdapterConfig, SchedulerConfig,
SpeculativeConfig)
ModelConfig, ParallelConfig, PromptAdapterConfig,
SchedulerConfig, SpeculativeConfig)
from vllm.executor.executor_base import ExecutorAsyncBase
from vllm.executor.gpu_executor import GPUExecutor
from vllm.logger import init_logger
@@ -29,7 +28,6 @@ class XPUExecutor(GPUExecutor):
device_config: DeviceConfig,
load_config: LoadConfig,
lora_config: Optional[LoRAConfig],
multimodal_config: Optional[MultiModalConfig],
prompt_adapter_config: Optional[PromptAdapterConfig],
speculative_config: Optional[SpeculativeConfig],
) -> None:
@@ -46,7 +44,6 @@ class XPUExecutor(GPUExecutor):
self.parallel_config = parallel_config
self.scheduler_config = scheduler_config
self.device_config = device_config
self.multimodal_config = multimodal_config
self.prompt_adapter_config = prompt_adapter_config
self.speculative_config = None