Move relevant docs to config

Author: murbans1

dockers/llm.rag.service/config.py

@@ -35,6 +35,7 @@ class WeaviateSettings(BaseSettings):
         default="my_custom_index",
         alias="WEAVIATE_INDEX_NAME",
     )
+    # TODO: consider moving to hybrid search config
     weaviate_hybrid_search_alpha: float = Field(
         default=0.5,
         alias="WEAVIATE_HYBRID_ALPHA",
@@ -43,12 +44,22 @@ class WeaviateSettings(BaseSettings):
         default="sentence-transformers/all-MiniLM-L6-v2",
         alias="EMBEDDING_MODEL_NAME",
     )
+    # TODO: consider moving to hybrid search config
+    hybrid_search_relevant_docs: Optional[int] = Field(
+        default=2,
+        alias="RELEVANT_DOCS",
+    )
 
     @field_validator("weaviate_hybrid_search_alpha", mode="before")
     @classmethod
     def validate_weaviate_hybrid_search_alpha(cls, v):
         return validate_float(v)
 
+    @field_validator("hybrid_search_relevant_docs", mode="before")
+    @classmethod
+    def validate_hybrid_search_relevant_docs(cls, v):
+        return validate_int(v)
+
     class Config:
         env_file = ".env"
         extra = "ignore"

dockers/llm.rag.service/serverragllm_csv_to_weaviate_local.py

@@ -43,7 +43,6 @@
 
 
 def setup(
-    relevant_docs: int,
     sql_search_db_and_model_path: str,
     max_context_length: int,
     sql_ticket_source: str,
@@ -86,7 +85,7 @@ def setup(
         vectorstore=vectorstore,
         client=client,
         system_prompt=SYSTEM_PROMPT_DEFAULT,
-        relevant_docs=relevant_docs,
+        relevant_docs=weaviate_settings.hybrid_search_relevant_docs,
         sql_search_db_and_model_path=sql_search_db_and_model_path,
         alpha=weaviate_settings.weaviate_hybrid_search_alpha,
         max_context_length=max_context_length,
@@ -105,12 +104,9 @@ def read_item(question: Union[str, None] = None):
 
 MICROSOFT_MODEL_ID = "microsoft/Phi-3-mini-4k-instruct"
 MOSAICML_MODEL_ID = "mosaicml/mpt-7b-chat"
-RELEVANT_DOCS_DEFAULT = 2
 SQL_SEARCH_DB_AND_MODEL_PATH_DEFAULT = "/app/db/"
 MODEL_MAX_CONTEXT_LEN = 8192
 
-relevant_docs = int(os.getenv("RELEVANT_DOCS", RELEVANT_DOCS_DEFAULT))
-
 sql_search_db_and_model_path = os.getenv(
     "SQL_SEARCH_DB_AND_MODEL_PATH", SQL_SEARCH_DB_AND_MODEL_PATH_DEFAULT
 )
@@ -122,7 +118,6 @@ def read_item(question: Union[str, None] = None):
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
 app = setup(
-    relevant_docs,
     sql_search_db_and_model_path,
     max_context_length,
     sql_ticket_source,