百度商业AI技术创新大赛-生成式推荐广告排序推理性能优化

Who this competition fits

适合已有推荐/广告排序模型经验，并能处理用户行为序列、广告文本与图片、上下文特征的开发者或小团队。参赛者还需要能用 AI Studio 账号完成实名认证；希望冲击线下赛的队伍应能准备可运行的推理脚本、环境构建脚本和优化后的模型文件。

What this competition gives you

Taken from the competition's official page.

• Free computeModel inference environment for entrants during recruitment
• DataAnonymized real user behavior and advertising dataset
• Support channelCompetition QQ question-and-answer group
• Support channelEmail support for competition questions
• Reference paperGRAB generative ranking model technical-practice paper
• Reference paperHSTU long-sequence behavior modeling paper

From registration to a first submission

1. 01

   Transformer 推理优化

2. 02

   广告推荐排序

3. 03

   Python 脚本开发

4. 04

   模型量化

5. 05

   Linux 环境构建

6. 06

   AUC 与 PCOC 指标

Before you commit: 难点在于同时守住排序效果与极低的端到端推理耗时：推理效率或策略效果任一项不达标，整体得分即为 0。你不能通过改动组网策略或用测试集训练来换分，还要把依赖安装、可能的编译步骤和推理流程封装进指定提交包，并满足评测容器的运行限制。