Today I'm publishing my new post about Kronk, this time talking about how it manages models, from how it handles them on disk to how it loads them sharing the memory between models.
Many interesting things happening about how it finds models directly in HuggingFace, or how it calculates the space needed for loading a model in your GPU.
Read it here: https://internals-for-interns.com/posts/kronk-model-management/








