feat(decorators): add performance measurement for streaming functions

This commit introduces a decorator `measure_stream_performance` that tracks and prints performance metrics for streaming functions. It measures time to the first character, total execution time, and calculates output rates for characters and tokens. Additionally, the `run_llm_stream` function now includes detailed documentation about its parameters and return type. These enhancements aim to facilitate better performance monitoring and debugging for developers working with streaming outputs in language models.

Author: huluxiaohuowa

hdl/utils/decorators/llm.py

@@ -2,6 +2,23 @@
 from functools import wraps
 
 def measure_stream_performance(func):
+    """
+    Measures the performance of a streaming function by tracking the time taken to output characters and the rate of character and token generation.
+
+    Args:
+        func (callable): The streaming function to be measured. It should return a generator that yields dictionaries containing 'type' and 'content' keys.
+
+    Returns:
+        callable: A wrapper function that performs the performance measurement and prints statistics related to the streaming output.
+
+    Statistics Printed:
+        - Time to first character (in seconds)
+        - Total time taken for the execution (in seconds)
+        - Total number of characters output
+        - Total number of tokens processed
+        - Characters output per second
+        - Tokens processed per second
+    """
     @wraps(func)
     def wrapper(*args, **kwargs):
         # 开始计时
@@ -54,6 +71,18 @@ def run_llm_stream(
     prompt,
     **kwargs
 ):
+    """
+    Run a language model stream with the given parameters.
+
+    Args:
+        llm (object): The language model object used to generate responses.
+        client_id (str): The unique identifier for the client making the request.
+        prompt (str): The input prompt to which the language model should respond.
+        **kwargs: Additional keyword arguments to customize the request.
+
+    Returns:
+        iterable: An iterable response stream from the language model.
+    """
     resp = llm.stream(
         client_id=client_id,
         prompt=prompt,