[vlm] Remove vision language config. (#6089)

Signed-off-by: Xiaowei Jiang <xwjiang2010@gmail.com>
Co-authored-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
xwjiang2010
2024-07-03 22:14:16 +00:00
committed by GitHub
co-authored by Roger Wang
parent 3c6325f0fc
commit d9e98f42e4
43 changed files with 371 additions and 465 deletions
+6 -7
View File
@@ -3,8 +3,8 @@ from abc import ABC, abstractmethod
from typing import List, Optional, Set, Tuple
from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
ModelConfig, ParallelConfig, SchedulerConfig,
SpeculativeConfig, VisionLanguageConfig)
ModelConfig, MultiModalConfig, ParallelConfig,
SchedulerConfig, SpeculativeConfig)
from vllm.lora.request import LoRARequest
from vllm.sequence import ExecuteModelRequest, SamplerOutput
@@ -26,7 +26,7 @@ class ExecutorBase(ABC):
device_config: DeviceConfig,
load_config: LoadConfig,
lora_config: Optional[LoRAConfig],
vision_language_config: Optional[VisionLanguageConfig],
multimodal_config: Optional[MultiModalConfig],
speculative_config: Optional[SpeculativeConfig],
) -> None:
self.model_config = model_config
@@ -36,7 +36,7 @@ class ExecutorBase(ABC):
self.parallel_config = parallel_config
self.scheduler_config = scheduler_config
self.device_config = device_config
self.vision_language_config = vision_language_config
self.multimodal_config = multimodal_config
self.speculative_config = speculative_config
self._init_executor()
@@ -120,7 +120,7 @@ class ExecutorAsyncBase(ExecutorBase):
device_config: DeviceConfig,
load_config: LoadConfig,
lora_config: Optional[LoRAConfig],
vision_language_config: Optional[VisionLanguageConfig],
multimodal_config: Optional[MultiModalConfig],
speculative_config: Optional[SpeculativeConfig],
) -> None:
# This locks each pipeline parallel stage so multiple virtual engines
@@ -132,8 +132,7 @@ class ExecutorAsyncBase(ExecutorBase):
super().__init__(model_config, cache_config, parallel_config,
scheduler_config, device_config, load_config,
lora_config, vision_language_config,
speculative_config)
lora_config, multimodal_config, speculative_config)
@abstractmethod
async def execute_model_async(