feat: Add token processing and generation speed to usage report

[sannysanoff]

Summary

Adds token speed metrics to the usage report when --show-speed flag is enabled.

Features

• --show-speed flag: Enable display of LLM timing and speed metrics (disabled by default)
• Time tracking: Measures total LLM response time and time to first token (TtFT)
• Speed calculation:
  • Prompt processing speed: prompt tokens / time to first token
  • Token generation speed: completion tokens / (total time - time to first token)

Output Example

With --show-speed enabled:

Tokens: 24k sent, 1.4k received.
LLM elapsed time: 35.00 seconds (TtFT: 2.50s)
Speed: 9600 prompt tokens/sec, 57 output tokens/sec

Without the flag, no timing information is shown.

Changes

• aider/args.py: Added --show-speed argument
• aider/coders/base_coder.py:
  • Track LLM response time after streaming completes
  • Track time to first token for streaming responses
  • Added _add_speed_info() helper method
  • Speed info only shown when --show-speed is enabled

🤖 Generated with Claude Code

[dwash96]

I'm not really into this structure of just sort of adding random if statements in an already long and dense file, instead of say a more general helpers/profiler.py file that, probably a start(self, type), log(self, type), elapsed(self, type) or similar. And the profiler handles gathering, and restoring the state of these types of timers. _add_speed_info() makes sense but I want to drive towards general utilities that handle discrete and useful operations

[sannysanoff]

Hi, thanks for taking time to review my slop. I tried to address your concerns and make it composable instead of adding complexity into already dense logic. Now base_coder.py diff is as minimal as possible, while keeping all functionality. Please advise if I can improve it even better.

[dwash96]

No problem, Time to first token is a good metric to have on hand, I think it's good as is, I'll change self.profiler to self.token_profiler after the merge but that's one regex replace lol

[dwash96]

Oh, but can you set up and run the pre-commit hook stuff, for the formatting? I want to be a bit more focused on making sure PRs are well formatted just as like a community standard

[sannysanoff]

Thanks, cool!