不记得这是个behavioral 还是technical问题.
behavioral的话就聊聊 what's the project bottleneck, revisit scope given the identified challenges, discuss priority trade-off with tl, etc
technical的话就聊聊how to improve model efficiency. 比如 data size, model size, parallelization multiple threading, lora, scaling law, 等等 有很多 vision language model 的efficiency新文章可以了解下