Move MultiModalConfig from config/__init__.py to config/multimodal.py (#24659)

Signed-off-by: Harry Mellor <[email protected]>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

Author: hmellor

tests/entrypoints/openai/test_lora_resolvers.py

@@ -9,7 +9,7 @@
 
 import pytest
 
-from vllm.config import MultiModalConfig
+from vllm.config.multimodal import MultiModalConfig
 from vllm.engine.multiprocessing.client import MQLLMEngineClient
 from vllm.entrypoints.openai.protocol import CompletionRequest, ErrorResponse
 from vllm.entrypoints.openai.serving_completion import OpenAIServingCompletion

tests/entrypoints/openai/test_serving_chat.py

@@ -12,7 +12,7 @@
 import pytest
 import pytest_asyncio
 
-from vllm.config import MultiModalConfig
+from vllm.config.multimodal import MultiModalConfig
 from vllm.engine.multiprocessing.client import MQLLMEngineClient
 from vllm.entrypoints.openai.protocol import ChatCompletionRequest
 from vllm.entrypoints.openai.serving_chat import OpenAIServingChat

tests/multimodal/test_cache.py

@@ -7,6 +7,7 @@
 import torch
 
 from vllm.config import ModelConfig, ParallelConfig, VllmConfig
+from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.cache import (MultiModalCache,
                                    MultiModalProcessorCacheItem,
                                    MultiModalProcessorCacheItemMetadata,
@@ -17,7 +18,6 @@
                                     MultiModalKwargsItems,
                                     MultiModalSharedField)
 from vllm.multimodal.processing import PromptInsertion
-from vllm.multimodal.registry import MultiModalRegistry
 
 
 def _dummy_elem(
@@ -96,7 +96,9 @@ def _create_vllm_config(
     enable_ipc: bool,
 ):
     return VllmConfig(
-        model_config=ModelConfig(mm_processor_cache_gb=mm_processor_cache_gb),
+        model_config=ModelConfig(
+            model="llava-hf/llava-onevision-qwen2-0.5b-ov-hf",
+            mm_processor_cache_gb=mm_processor_cache_gb),
         parallel_config=ParallelConfig(
             data_parallel_size=1 if enable_ipc else 2),
     )
@@ -113,15 +115,16 @@ def _compare_caches(
     n_iter: int = 100,
     seed: int = 0,
 ):
-    mm_registry = MultiModalRegistry()
-    cache_0_p0 = processor_cache_from_config(config_0, mm_registry)
-    cache_0_p1 = engine_receiver_cache_from_config(config_0, mm_registry)
-    cache_1_p0 = processor_cache_from_config(config_1, mm_registry)
-    cache_1_p1 = engine_receiver_cache_from_config(config_1, mm_registry)
+    cache_0_p0 = processor_cache_from_config(config_0, MULTIMODAL_REGISTRY)
+    cache_0_p1 = engine_receiver_cache_from_config(config_0,
+                                                   MULTIMODAL_REGISTRY)
+    cache_1_p0 = processor_cache_from_config(config_1, MULTIMODAL_REGISTRY)
+    cache_1_p1 = engine_receiver_cache_from_config(config_1,
+                                                   MULTIMODAL_REGISTRY)
 
     cache_size_gb = max(
-        config_0.model_config.mm_processor_cache_gb,
-        config_1.model_config.mm_processor_cache_gb,
+        config_0.model_config.multimodal_config.mm_processor_cache_gb,
+        config_1.model_config.multimodal_config.mm_processor_cache_gb,
     )
     item_size_gb = int(cache_size_gb / item_capacity)
 

tests/test_config.py

@@ -6,9 +6,9 @@
 import pytest
 
 from vllm.compilation.backends import VllmBackend
-from vllm.config import (ModelConfig, PoolerConfig, VllmConfig, get_field,
-                         update_config)
+from vllm.config import ModelConfig, PoolerConfig, VllmConfig, update_config
 from vllm.config.load import LoadConfig
+from vllm.config.utils import get_field
 from vllm.model_executor.layers.pooler import PoolingType
 from vllm.platforms import current_platform
 

tests/v1/engine/test_processor_multi_modal_uuids.py

@@ -31,7 +31,7 @@ def _mk_processor(monkeypatch,
                         raising=True)
     monkeypatch.setattr(ModelConfig,
                         "__post_init__",
-                        lambda self: None,
+                        lambda self, *args: None,
                         raising=True)
     monkeypatch.setattr(UnspecifiedPlatform,
                         "is_async_output_supported",