Doc: update description for efficiency test

[anyangml]

This PR updates the description for efficiency test. The changes were made in #264.

[Copilot]

Pull Request Overview

This PR revises the efficiency test description to reflect updated sample counts, warm-up phase, and dynamic GPU utilization.

• Reduced the number of frames from 2000 to 1000 and adjusted the warm-up phase from 20% to 10%.
• Added a binary search step for dynamically sizing each frame to fully utilize GPU capacity.
• Updated formulas and denominators to match the new 900-configuration average.

[Copilot]

[nitpick] Consider adding a brief reference or link to where the binary search algorithm is implemented, or a note on its termination criteria, to improve reproducibility.

Suggested change

To assess the efficiency of the model, we randomly selected 1000 frames from the domain of Inorganic Materials and Catalysis using the aforementioned out-of-distribution datasets. Each frame was expanded to contain between 800 and 1000 atoms — dynamically determined using a binary search algorithm to fully utilize GPU capacity — by replicating the unit cell. This ensured that measurements of inference efficiency were conducted within the regime of convergence. The initial 10% of the test samples were considered a warm-up phase and thus were excluded from the efficiency timing. We have reported the average efficiency across the remaining 900 frames.

To assess the efficiency of the model, we randomly selected 1000 frames from the domain of Inorganic Materials and Catalysis using the aforementioned out-of-distribution datasets. Each frame was expanded to contain between 800 and 1000 atoms — dynamically determined using a binary search algorithm to fully utilize GPU capacity — by replicating the unit cell. The binary search algorithm iteratively adjusts the number of atoms until the GPU capacity is maximized, terminating when the difference between the estimated and actual GPU utilization falls below a predefined threshold. For implementation details, refer to [Binary Search Algorithm Documentation](https://example.com/binary-search-doc). This ensured that measurements of inference efficiency were conducted within the regime of convergence. The initial 10% of the test samples were considered a warm-up phase and thus were excluded from the efficiency timing. We have reported the average efficiency across the remaining 900 frames.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 65.54%. Comparing base (9fadf6d) to head (8d2464a).
Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #304   +/-   ##
=======================================
  Coverage   65.54%   65.54%           
=======================================
  Files          35       35           
  Lines        1550     1550           
  Branches      185      185           
=======================================
  Hits         1016     1016           
  Misses        496      496           
  Partials       38       38           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.