Revert "[Renderer] Separate out RendererConfig from ModelConfig (#30145)" (#30199)

This commit is contained in:
Cyrus Leung
2025-12-07 00:00:22 -08:00
committed by GitHub
parent 27f4c2fd46
commit e83b7e379c
105 changed files with 797 additions and 969 deletions
+7 -12
View File
@@ -7,7 +7,7 @@ import torch
from vllm.attention.backends.abstract import AttentionBackend
from vllm.attention.layer import Attention
from vllm.config import RendererConfig, SchedulerConfig, VllmConfig
from vllm.config import ModelConfig, SchedulerConfig, VllmConfig
from vllm.model_executor.models.interfaces import MultiModalEmbeddings
from vllm.model_executor.models.utils import extract_layer_index
from vllm.multimodal.cache import processor_only_cache_from_config
@@ -23,29 +23,24 @@ class MultiModalBudget:
def __init__(
self,
renderer_config: RendererConfig,
model_config: ModelConfig,
scheduler_config: SchedulerConfig,
mm_registry: MultiModalRegistry,
) -> None:
super().__init__()
self.renderer_config = renderer_config
self.model_config = renderer_config.model_config
self.model_config = model_config
self.scheduler_config = scheduler_config
self.mm_registry = mm_registry
self.cache = cache = processor_only_cache_from_config(
renderer_config, mm_registry
)
self.cache = cache = processor_only_cache_from_config(model_config, mm_registry)
self.max_model_len = self.model_config.max_model_len
self.max_model_len = model_config.max_model_len
self.max_num_reqs = scheduler_config.max_num_seqs
self.mm_limits = mm_registry.get_mm_limits_per_prompt(
renderer_config, cache=cache
)
self.mm_limits = mm_registry.get_mm_limits_per_prompt(model_config, cache=cache)
max_tokens_by_modality = mm_registry.get_max_tokens_per_item_by_modality(
renderer_config,
model_config,
cache=cache,
profiler_limits=self.mm_limits,
)