About the Challenge：In this challenge, teams must leverage the heterogeneous computing power of AI PCs to train models. Their objective is to continuously explore the Honor of Kings 1v1 map to learn the optimal strategy for victory, aiming to be the first to destroy the opponent's crystal.

The map used in this challenge is elongated, with the resurrection points of both intelligences at either end and their respective camp's crystal located in front of these points. The crystals continuously spawn minions for their camp, which automatically advance toward the enemy camp, attacking any defense towers, crystals, and heroes along their path. Defense towers located in front of the crystals can attack enemy heroes and minions that come within range. The intelligences have the freedom to move anywhere on the map and use their abilities. (A comprehensive development guide will be provided on the platform following successful registration.)

Objective：Within a designated timeframe, fully utilize the heterogeneous computing power of AI PCs to train models, enabling them to continuously explore the 1v1 map and learn the optimal winning strategy.

Points Rules：In each match between two teams, all heroes from both sides must compete, facing each hero from the opposing team in turn, with a total of two matches as sides are swapped. A win grants 1 point, with no points awarded for a loss.

Ranking Rules：In the tournament's final week, the system will automatically initiate matches using the latest models submitted by the participating teams. The exact number of rounds for leaderboard matches will be decided later, and rankings will be determined by the total points accumulated by each team.

AIPC High Performance Gaming Track Setting Comprehensive Ranking and Efficiency Ranking.

The comprehensive ranking directly uses the models submitted by each participating team for competition, evaluates the maximum model capability that players can achieve by using Intel AI PC terminals for local training during the competition, and ranks first, second, and third place.

The efficiency ranking uses training code packages submitted by each participating team for a fixed duration of hosted training. The last model checkpoint before the end of the training is taken for competition, in order to evaluate the performance of the participating teams in engineering optimization for the Intel AI PC platform and award the Best Efficiency Award.

Note: In order to minimize the influence of model size and theoretical computational complexity, the efficiency ranking code package hosting training output model checkpoints should be equivalent to the number of model parameters corresponding to the initial code package, with an error not exceeding 5%.