[VLM] Refactor MultiModalConfig initialization and profiling (#7530)
This commit is contained in:
@@ -7,7 +7,7 @@ from typing import (TYPE_CHECKING, Dict, List, Mapping, Optional, Tuple, Type,
|
||||
|
||||
from vllm.config import (CacheConfig, DecodingConfig, DeviceConfig,
|
||||
EngineConfig, LoadConfig, LoRAConfig, ModelConfig,
|
||||
MultiModalConfig, ObservabilityConfig, ParallelConfig,
|
||||
ObservabilityConfig, ParallelConfig,
|
||||
PromptAdapterConfig, SchedulerConfig,
|
||||
SpeculativeConfig, TokenizerPoolConfig)
|
||||
from vllm.executor.executor_base import ExecutorBase
|
||||
@@ -765,9 +765,6 @@ class EngineArgs:
|
||||
"CPU offload space must be non-negative"
|
||||
f", but got {self.cpu_offload_gb}")
|
||||
|
||||
multimodal_config = MultiModalConfig(
|
||||
limit_per_prompt=self.limit_mm_per_prompt or {})
|
||||
|
||||
device_config = DeviceConfig(device=self.device)
|
||||
model_config = ModelConfig(
|
||||
model=self.model,
|
||||
@@ -791,7 +788,8 @@ class EngineArgs:
|
||||
disable_sliding_window=self.disable_sliding_window,
|
||||
skip_tokenizer_init=self.skip_tokenizer_init,
|
||||
served_model_name=self.served_model_name,
|
||||
multimodal_config=multimodal_config)
|
||||
limit_mm_per_prompt=self.limit_mm_per_prompt,
|
||||
)
|
||||
cache_config = CacheConfig(
|
||||
block_size=self.block_size,
|
||||
gpu_memory_utilization=self.gpu_memory_utilization,
|
||||
@@ -970,7 +968,6 @@ class EngineArgs:
|
||||
scheduler_config=scheduler_config,
|
||||
device_config=device_config,
|
||||
lora_config=lora_config,
|
||||
multimodal_config=multimodal_config,
|
||||
speculative_config=speculative_config,
|
||||
load_config=load_config,
|
||||
decoding_config=decoding_config,
|
||||
|
||||
@@ -10,7 +10,7 @@ from typing_extensions import assert_never
|
||||
import vllm.envs as envs
|
||||
from vllm.config import (CacheConfig, DecodingConfig, DeviceConfig,
|
||||
EngineConfig, LoadConfig, LoRAConfig, ModelConfig,
|
||||
MultiModalConfig, ObservabilityConfig, ParallelConfig,
|
||||
ObservabilityConfig, ParallelConfig,
|
||||
PromptAdapterConfig, SchedulerConfig,
|
||||
SpeculativeConfig)
|
||||
from vllm.core.scheduler import (ScheduledSequenceGroup, Scheduler,
|
||||
@@ -100,8 +100,6 @@ class LLMEngine:
|
||||
scheduler_config: The configuration related to the request scheduler.
|
||||
device_config: The configuration related to the device.
|
||||
lora_config (Optional): The configuration related to serving multi-LoRA.
|
||||
multimodal_config (Optional): The configuration related to multimodal
|
||||
models.
|
||||
speculative_config (Optional): The configuration related to speculative
|
||||
decoding.
|
||||
executor_class: The model executor class for managing distributed
|
||||
@@ -172,7 +170,6 @@ class LLMEngine:
|
||||
device_config: DeviceConfig,
|
||||
load_config: LoadConfig,
|
||||
lora_config: Optional[LoRAConfig],
|
||||
multimodal_config: Optional[MultiModalConfig],
|
||||
speculative_config: Optional[SpeculativeConfig],
|
||||
decoding_config: Optional[DecodingConfig],
|
||||
observability_config: Optional[ObservabilityConfig],
|
||||
@@ -235,7 +232,6 @@ class LLMEngine:
|
||||
self.model_config = model_config
|
||||
self.cache_config = cache_config
|
||||
self.lora_config = lora_config
|
||||
self.multimodal_config = multimodal_config
|
||||
self.parallel_config = parallel_config
|
||||
self.scheduler_config = scheduler_config
|
||||
self.device_config = device_config
|
||||
@@ -278,7 +274,6 @@ class LLMEngine:
|
||||
scheduler_config=scheduler_config,
|
||||
device_config=device_config,
|
||||
lora_config=lora_config,
|
||||
multimodal_config=multimodal_config,
|
||||
speculative_config=speculative_config,
|
||||
load_config=load_config,
|
||||
prompt_adapter_config=prompt_adapter_config,
|
||||
|
||||
Reference in New Issue
Block a user