Use weaviate config

Author: murbans1

dockers/llm.rag.service/config.py

@@ -1,8 +1,7 @@
-from pydantic_settings import BaseSettings
-from pydantic import Field
-
 from typing import Optional
 
+from pydantic import Field, field_validator
+from pydantic_settings import BaseSettings
 
 WEAVIATE_HYBRID_ALPHA_DEFAULT = 0.5
 
@@ -11,7 +10,7 @@ def validate_float(value):
     if type(value) == float:
         return value
     try:
-        return float(value.strip("'").strip("\""))
+        return float(value.strip("'").strip('"'))
     except (TypeError, ValueError):
         raise ValueError("Value must be convertible to a float")
 
@@ -34,9 +33,28 @@ class WeaviateSettings(BaseSettings):
         alias="WEAVIATE_HYBRID_ALPHA",
     )
 
+    @field_validator("weaviate_hybrid_search_alpha", mode="before")
+    @classmethod
+    def validate_weaviate_hybrid_search_alpha(cls, v):
+        return validate_float(v)
+
     class Config:
         env_file = ".env"
         extra = "ignore"
 
     def is_set(self) -> bool:
-        return all([self.weaviate_uri, self.weaviate_grpc_uri, self.weaviate_index_name])
+        return all(
+            [self.weaviate_uri, self.weaviate_grpc_uri, self.weaviate_index_name]
+        )
+
+    def get_weaviate_uri(self):
+        return self.weaviate_uri.split(":")[0]
+
+    def get_weaviate_port(self):
+        return int(self.weaviate_uri.split(":")[1])
+
+    def get_weaviate_grpc_uri(self):
+        return self.weaviate_grpc_uri.split(":")[0]
+
+    def get_weaviate_grpc_port(self):
+        return int(self.weaviate_grpc_uri.split(":")[1])

dockers/llm.rag.service/search.py

@@ -29,6 +29,7 @@
 
 
 from common import get_answer_with_settings_with_weaviate_filter
+from config import WeaviateSettings
 
 SYSTEM_PROMPT_DEFAULT = """You are a specialized support ticket assistant. Format your responses following these rules:
                 1. Answer the provided question only using the provided context.
@@ -47,31 +48,29 @@ def setup(
     model_id: str,
     max_tokens: int,
     model_temperature: float,
-    weaviate_url: str,
-    weaviate_grpc_url: str,
-    weaviate_index: str,
     embedding_model_name: str,
     sql_search_db_and_model_path: str,
-    alpha: float,
     max_context_length: int,
     sql_ticket_source: str,
 ):
     app = FastAPI()
 
+    weaviate_settings = WeaviateSettings()
+
     embeddings = HuggingFaceEmbeddings(model_name=embedding_model_name)
 
     weaviate_client = weaviate.connect_to_custom(
-        http_host=weaviate_url.split(":")[0],
-        http_port=int(weaviate_url.split(":")[1]),
+        http_host=weaviate_settings.get_weaviate_uri(),
+        http_port=weaviate_settings.get_weaviate_port(),
         http_secure=False,
-        grpc_host=weaviate_grpc_url.split(":")[0],
-        grpc_port=int(weaviate_grpc_url.split(":")[1]),
+        grpc_host=weaviate_settings.get_weaviate_grpc_uri(),
+        grpc_port=weaviate_settings.get_weaviate_grpc_port(),
         grpc_secure=False,
     )
 
     vectorstore = WeaviateVectorStore(
         client=weaviate_client,
-        index_name=weaviate_index,
+        index_name=weaviate_settings.weaviate_index_name,
         text_key="text",
         embedding=embeddings,
     )
@@ -98,7 +97,7 @@ def setup(
         relevant_docs=relevant_docs,
         llm_server_url=llm_server_url,
         sql_search_db_and_model_path=sql_search_db_and_model_path,
-        alpha=alpha,
+        alpha=weaviate_settings.weaviate_hybrid_search_alpha,
         max_context_length=max_context_length,
         sql_ticket_source=sql_ticket_source,
     )
@@ -161,12 +160,8 @@ def read_item(question: Union[str, None] = None):
     model_id,
     max_tokens,
     model_temperature,
-    weaviate_url,
-    weaviate_grpc_url,
-    weaviate_index,
     embedding_model_name,
     sql_search_db_and_model_path,
-    alpha,
     max_context_length,
     sql_ticket_source,
 )