469: agent profiler, waterfall chart, step summary table

[paulpalmieri]

Notebook usage only for now:

[paulpalmieri]

@adeprez would you be available to discuss implementation ? I've used globals to be able to decorate any functions but ideally I would avoid that.

[paulpalmieri]

Important: currently limited to Jupyter notebooks as it returns a plot and a dataframe.

Usage

Use this code after an agent completes.

plot, table = agent.get_summary()

To decorate a function:

from lavague.core.utilities.profiling import time_profiler

with time_profiler("event_name"):
    your_function_call()

Currently tracked steps:

• world model inference
• retrieval
• navigation engine inference
• execute code