在不同训练任务中在图片粒度上共享隐空间缓存 Sharing latents cache between different training tasks at the image level

[astro-jingtao]

当前设置 cache_latents: True 似乎只能在本次训练中使用缓存，或者当训练数据完全一样的时候可以通过设置 cache_path 来复用缓存。是否有计划支持在图片粒度上复用缓存？

The current configuration cache_latents: True seems to only enable caching within the current training session, or to reuse the cache by setting cache_path when the training data is exactly the same. Are there any plans to support cache reuse at the image level?