Merge pull request road-core#58 from tisnik/minor-fixes

Minor fixes in service source code

ols/app/endpoints/health.py

@@ -16,7 +16,7 @@
 
 router = APIRouter(tags=["health"])
 logger = logging.getLogger(__name__)
-llm_is_ready_persistent_state = False
+llm_is_ready_persistent_state: bool = False
 
 
 def llm_is_ready() -> bool:
@@ -25,7 +25,7 @@ def llm_is_ready() -> bool:
     If so, store the success to `llm_is_ready_persistent_state` to cache
     the result for future calls.
     """
-    global llm_is_ready_persistent_state
+    global llm_is_ready_persistent_state  # pylint: disable=global-statement
     if llm_is_ready_persistent_state is True:
         return True
     try:
@@ -40,7 +40,7 @@ def llm_is_ready() -> bool:
             return True
         raise ValueError(f"Unexpected response from LLM: {response}")
     except Exception as e:
-        logger.error(f"LLM connection check failed with - {e}")
+        logger.error("LLM connection check failed with - %s", e)
         return False
 
 

ols/app/main.py

@@ -119,7 +119,9 @@ async def stream_response_body(
             )
             yield chunk
 
-    if isinstance(response, StreamingResponse):
+    # current version of Starlette pass instance of _StreamingResponse class that is
+    # private. Thus we need to check if the body_iterator attribute exists
+    if hasattr(response, "body_iterator"):
         # The response is already a streaming response
         response.body_iterator = stream_response_body(response.body_iterator)
     else:

ols/app/models/config.py

@@ -540,7 +540,7 @@ def __init__(self, **data: Any) -> None:
         super().__init__(**data)
         # password should be read from file
         if self.password_path is not None:
-            with open(self.password_path) as f:
+            with open(self.password_path, "r") as f:
                 self.password = f.read().rstrip()
 
     @model_validator(mode="after")

ols/src/llms/llm_loader.py

@@ -89,7 +89,7 @@ def load_llm(
             f"Unsupported LLM provider type '{provider_config.type}'."
         )
 
-    logger.debug(f"loading LLM '{model}' from '{provider}'")
+    logger.debug("loading LLM model '%s' from provider '%s'", model, provider)
 
     llm_provider = llm_providers_reg.llm_providers[provider_config.type]
     return llm_provider(model, provider_config, generic_llm_params or {}).load()

ols/src/prompts/prompts.py

@@ -5,8 +5,6 @@
 
 from ols.constants import SUBJECT_ALLOWED, SUBJECT_REJECTED
 
-# TODO: OLS-503 Fine tune system prompt
-
 # Note::
 # Right now templates are somewhat alligned to make granite work better.
 # GPT still works well with this. Ideally we should have model specific tags.