Multimodal: Fix memory leak with MMEmbeddings

On a basic python class, class attributes are handled by reference,
meaning that every instance of embeddings would attach to that reference
and allocate more memory.

Switch to a Pydantic class and factory methods when instantiating.

Signed-off-by: kingbri <[email protected]>

common/multimodal.py

@@ -1,20 +1,20 @@
-from typing import List
 from backends.exllamav2.vision import get_image_embedding
 from common import model
 from loguru import logger
+from pydantic import BaseModel, Field
+from typing import List
 
 from common.optional_dependencies import dependencies
 
 if dependencies.exllamav2:
     from exllamav2 import ExLlamaV2VisionTower
 
-
-class MultimodalEmbeddingWrapper:
+class MultimodalEmbeddingWrapper(BaseModel):
     """Common multimodal embedding wrapper"""
 
     type: str = None
-    content: List = []
-    text_alias: List[str] = []
+    content: list = Field(default_factory=list)
+    text_alias: List[str] = Field(default_factory=list)
 
     async def add(self, url: str):
         # Determine the type of vision embedding to use