feat: current usage #821
feat: current usage #821
Conversation
Summary of ChangesHello @suluyana, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request implements a comprehensive token usage tracking mechanism for the LLM agent. It introduces global counters for prompt and completion tokens, protected by an asynchronous lock, to provide an aggregated view of token consumption throughout the application's lifecycle. Additionally, it ensures that usage data is consistently captured from OpenAI API calls, including streaming responses, and logs this information at each step of the agent's operation. Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces token usage tracking for LLM calls, accumulating prompt and completion tokens at a process level. The implementation in llm_agent.py uses global variables for tracking, which I've recommended refactoring into class attributes on LLMAgent to improve encapsulation and testability. I've provided specific suggestions on how to implement this change. Additionally, I've suggested a minor readability improvement in openai_llm.py for the logic that enables usage data in streaming responses.
| # Current process shared | ||
| TOTAL_PROMPT_TOKENS = 0 | ||
| TOTAL_COMPLETION_TOKENS = 0 | ||
| TOKEN_LOCK = asyncio.Lock() |
There was a problem hiding this comment.
Using global variables for process-wide state can lead to code that is hard to test and maintain. It's better to encapsulate this state within the LLMAgent class itself as class attributes. This clearly associates the state with the agent and avoids polluting the global namespace.
For example, you could define them inside LLMAgent like this:
class LLMAgent(Agent):
TOTAL_PROMPT_TOKENS = 0
TOTAL_COMPLETION_TOKENS = 0
TOKEN_LOCK = asyncio.Lock()
...| prompt_tokens = _response_message.prompt_tokens | ||
| completion_tokens = _response_message.completion_tokens | ||
|
|
||
| global TOTAL_PROMPT_TOKENS, TOTAL_COMPLETION_TOKENS, TOKEN_LOCK | ||
| async with TOKEN_LOCK: | ||
| TOTAL_PROMPT_TOKENS += prompt_tokens | ||
| TOTAL_COMPLETION_TOKENS += completion_tokens | ||
|
|
||
| # tokens in the current step | ||
| self.log_output( | ||
| f'[usage] prompt_tokens: {prompt_tokens}, completion_tokens: {completion_tokens}' | ||
| ) | ||
| # total tokens for the process so far | ||
| self.log_output( | ||
| f'[usage] prompt_tokens: {_response_message.prompt_tokens}, ' | ||
| f'completion_tokens: {_response_message.completion_tokens}') | ||
| f'[usage_total] total_prompt_tokens: {TOTAL_PROMPT_TOKENS}, ' | ||
| f'total_completion_tokens: {TOTAL_COMPLETION_TOKENS}') | ||
|
|
There was a problem hiding this comment.
Following my suggestion to move the tracking variables to be class attributes of LLMAgent, this block should be updated to access them via the class. This avoids the global statement and makes it clear that you're modifying the shared state of the LLMAgent class.
prompt_tokens = _response_message.prompt_tokens
completion_tokens = _response_message.completion_tokens
# 使用全局累积
async with LLMAgent.TOKEN_LOCK:
LLMAgent.TOTAL_PROMPT_TOKENS += prompt_tokens
LLMAgent.TOTAL_COMPLETION_TOKENS += completion_tokens
# tokens in the current step
self.log_output(
f'[usage] prompt_tokens: {prompt_tokens}, completion_tokens: {completion_tokens}'
)
# total tokens for the process so far
self.log_output(
f'[usage_total] total_prompt_tokens: {LLMAgent.TOTAL_PROMPT_TOKENS}, '
f'total_completion_tokens: {LLMAgent.TOTAL_COMPLETION_TOKENS}')| if kwargs.get('stream', False) and self.args.get( | ||
| 'stream_options', {}).get('include_usage', True): | ||
| kwargs.setdefault('stream_options', {})['include_usage'] = True |
There was a problem hiding this comment.
This conditional logic is a bit dense and could be hard to parse. For better readability, you could break it down into a few lines with intermediate variables and a comment explaining the intent.
| if kwargs.get('stream', False) and self.args.get( | |
| 'stream_options', {}).get('include_usage', True): | |
| kwargs.setdefault('stream_options', {})['include_usage'] = True | |
| is_streaming = kwargs.get('stream', False) | |
| stream_options_config = self.args.get('stream_options', {}) | |
| # For streaming responses, we should request usage statistics by default, | |
| # unless it's explicitly disabled in the configuration. | |
| if is_streaming and stream_options_config.get('include_usage', True): | |
| kwargs.setdefault('stream_options', {})['include_usage'] = True |
1 participant
Change Summary
Related issue number
Checklist
pre-commit installandpre-commit run --all-filesbefore git commit, and passed lint check.