Added more detailed metrics to logger that help characterize an LLM's…

… performance.

phi/llm/ollama/chat.py

@@ -270,7 +270,7 @@ def response_stream(self, messages: List[Message]) -> Iterator[str]:
  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}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