Added adapter_instance_id and run_id (#40)

* Added adapter_instance_id and run_id

* Updated field model_type -> model_name

* Review comment fixes

* Version bump

---------

Co-authored-by: Rahul Johny <[email protected]>

Author: Deepak-Kesavan

src/unstract/sdk/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.22.1"
+__version__ = "0.23.0"
 
 
 def get_sdk_version():

src/unstract/sdk/audit.py

@@ -15,15 +15,6 @@ class Audit(StreamMixin):
 
     Attributes:
         None
-
-    Example usage:
-        audit = Audit()
-        audit.push_usage_data(
-            token_counter,
-            workflow_id,
-            execution_id,
-            external_service,
-            event_type)
     """
 
     def __init__(self, log_level: LogLevel = LogLevel.INFO) -> None:
@@ -33,23 +24,28 @@ def push_usage_data(
         self,
         platform_api_key: str,
         token_counter: TokenCountingHandler = None,
-        workflow_id: str = "",
-        execution_id: str = "",
-        external_service: str = "",
+        model_name: str = "",
         event_type: CBEventType = None,
+        **kwargs,
     ) -> None:
         """Pushes the usage data to the platform service.
 
         Args:
+            platform_api_key (str): The platform API key.
             token_counter (TokenCountingHandler, optional): The token counter
-              object. Defaults to None.
-            workflow_id (str, optional): The ID of the workflow. Defaults to "".
-            execution_id (str, optional): The ID of the execution. Defaults
-              to "".
-            external_service (str, optional): The name of the external service.
-              Defaults to "".
+                object. Defaults to None.
+            model_name (str, optional): The name of the model.
+                Defaults to "".
             event_type (CBEventType, optional): The type of the event. Defaults
-              to None.
+                to None.
+            **kwargs: Optional keyword arguments.
+                workflow_id (str, optional): The ID of the workflow.
+                    Defaults to "".
+                execution_id (str, optional): The ID of the execution. Defaults
+                    to "".
+                adapter_instance_id (str, optional): The adapter instance ID.
+                    Defaults to "".
+                run_id (str, optional): The run ID. Defaults to "".
 
         Returns:
             None
@@ -66,11 +62,18 @@ def push_usage_data(
         )
         bearer_token = platform_api_key
 
+        workflow_id = kwargs.get("workflow_id", "")
+        execution_id = kwargs.get("execution_id", "")
+        adapter_instance_id = kwargs.get("adapter_instance_id", "")
+        run_id = kwargs.get("run_id", "")
+
         data = {
-            "usage_type": event_type,
-            "external_service": external_service,
             "workflow_id": workflow_id,
             "execution_id": execution_id,
+            "adapter_instance_id": adapter_instance_id,
+            "run_id": run_id,
+            "usage_type": event_type,
+            "model_name": model_name,
             "embedding_tokens": token_counter.total_embedding_token_count,
             "prompt_tokens": token_counter.prompt_llm_token_count,
             "completion_tokens": token_counter.completion_llm_token_count,
@@ -100,3 +103,6 @@ def push_usage_data(
                 log=f"Error while pushing usage details: {e}",
                 level=LogLevel.ERROR,
             )
+
+        finally:
+            token_counter.reset_counts()

src/unstract/sdk/llm.py

@@ -57,32 +57,26 @@ def run_completion(
     ) -> Optional[dict[str, Any]]:
         # Setup callback manager to collect Usage stats
         UNCallbackManager.set_callback_manager(
-            platform_api_key=platform_api_key, llm=llm
+            platform_api_key=platform_api_key, llm=llm, **kwargs
         )
+        # Removing specific keys from kwargs
+        new_kwargs = kwargs.copy()
+        for key in [
+            "workflow_id",
+            "execution_id",
+            "adapter_instance_id",
+            "run_id",
+        ]:
+            new_kwargs.pop(key, None)
         for i in range(retries):
             try:
-                response: CompletionResponse = llm.complete(prompt, **kwargs)
+                response: CompletionResponse = llm.complete(
+                    prompt, **new_kwargs
+                )
                 match = cls.json_regex.search(response.text)
                 if match:
                     response.text = match.group(0)
-
-                usage = {}
-                llm_token_counts = llm.callback_manager.handlers[
-                    0
-                ].llm_token_counts
-                if llm_token_counts:
-                    llm_token_count = llm_token_counts[0]
-                    usage[
-                        "prompt_token_count"
-                    ] = llm_token_count.prompt_token_count
-                    usage[
-                        "completion_token_count"
-                    ] = llm_token_count.completion_token_count
-                    usage[
-                        "total_token_count"
-                    ] = llm_token_count.total_token_count
-
-                return {"response": response, "usage": usage}
+                return {"response": response}
 
             except Exception as e:
                 if i == retries - 1:

src/unstract/sdk/utils/callback_manager.py

@@ -41,8 +41,7 @@ def set_callback_manager(
         platform_api_key: str,
         llm: Optional[LLM] = None,
         embedding: Optional[BaseEmbedding] = None,
-        workflow_id: str = "",
-        execution_id: str = "",
+        **kwargs,
     ) -> LlamaIndexCallbackManager:
         """Sets the standard callback manager for the llm. This is to be called
         explicitly whenever there is a need for the callback handling defined
@@ -52,7 +51,7 @@ def set_callback_manager(
             llm (LLM): The LLM type
 
         Returns:
-            CallbackManager tyoe of llama index
+            CallbackManager type of llama index
 
         Example:
             UNCallbackManager.set_callback_manager(
@@ -73,8 +72,7 @@ def set_callback_manager(
             platform_api_key=platform_api_key,
             llm_model=llm,
             embed_model=embedding,
-            workflow_id=workflow_id,
-            execution_id=execution_id,
+            **kwargs,
         )
 
         callback_manager: LlamaIndexCallbackManager = LlamaIndexCallbackManager(

src/unstract/sdk/utils/usage_handler.py

@@ -36,16 +36,14 @@ def __init__(
         platform_api_key: str,
         llm_model: LLM = None,
         embed_model: BaseEmbedding = None,
-        workflow_id: str = "",
-        execution_id: str = "",
         event_starts_to_ignore: Optional[list[CBEventType]] = None,
         event_ends_to_ignore: Optional[list[CBEventType]] = None,
         verbose: bool = False,
         log_level: LogLevel = LogLevel.INFO,
+        **kwargs,
     ) -> None:
+        self.kwargs = kwargs
         self._verbose = verbose
-        self.workflow_id = workflow_id
-        self.execution_id = execution_id
         self.token_counter = token_counter
         self.llm_model = llm_model
         self.embed_model = embed_model
@@ -96,9 +94,8 @@ def on_event_end(
                 platform_api_key=self.platform_api_key,
                 token_counter=self.token_counter,
                 event_type=event_type,
-                external_service=self.llm_model.metadata.model_name,
-                workflow_id=self.workflow_id,
-                execution_id=self.execution_id,
+                model_name=self.llm_model.metadata.model_name,
+                **self.kwargs,
             )
 
         elif (
@@ -113,7 +110,6 @@ def on_event_end(
                 platform_api_key=self.platform_api_key,
                 token_counter=self.token_counter,
                 event_type=event_type,
-                external_service=self.embed_model.model_name,
-                workflow_id=self.workflow_id,
-                execution_id=self.execution_id,
+                model_name=self.embed_model.model_name,
+                **self.kwargs,
             )