Merge pull request phidatahq#217 from shailensobhee/main

More detailed LLM performance metrics added to Logger.
utopic-dev · May 5, 2024 · bb74a4a · bb74a4a
2 parents a6db51c + 4101843
commit bb74a4a
Showing 1 changed file with 3 additions and 0 deletions.
diff --git a/phi/llm/ollama/chat.py b/phi/llm/ollama/chat.py
@@ -268,6 +268,9 @@ def response_stream(self, messages: List[Message]) -> Iterator[str]:
  yield response_content
 
  response_timer.stop()
+ logger.debug(f"Number of tokens generated: {completion_tokens}")
+ logger.debug(f"Time per output token: {response_timer.elapsed/completion_tokens:.4f}s")
+ logger.debug(f"Throughtput: {completion_tokens/response_timer.elapsed:.4f} tokens/s")
  logger.debug(f"Time to generate response: {response_timer.elapsed:.4f}s")
 
  # -*- Create assistant message