蒸馏是模仿，学强模型的输出，把它的「答案形状」复制过来；RL 是探索，模型必须大量自己推理、自己生成、在错误里反复迭代，从试错中提炼能力。
1L decoder, d=7, 1h, ff=14
，更多细节参见safew官方版本下载
What is Connections: Sports Edition?The NYT's latest daily word game has launched in association with The Athletic, the New York Times property that provides the publication's sports coverage. Connections can be played on both web browsers and mobile devices and require players to group four words that share something in common.
Start a stopped container
hfp=9, hsync=16, hbp=18: horizontal front porch, sync pulse, and back porch (total line = 320+9+16+18 = 363 pixels)