M. Yusuf Sarıgöz

Hey @Yossef-Dawoad, with #75, now we're using GGUF and older model files are not usable. can you please update the notebook?

Batch inference is working on the C/C++ side, it needs to be exposed to Python. I still have some considerations about how to expose it. The issue is, if we...

It requests ~12 mb instead of 8mb that I set as a fixed value here. https://github.com/monatis/clip.cpp/blob/e2eee8e9b11afe4fc9fdb22d1f6d0ea53df9552a/clip.cpp#L24-L30 You can slightly increase them --8 is for patch32 and 16 is for patch16,...

> That worked! That's great! I'll try to find the root cause of this difference and patch it later on. > Not very low-end in the grand scheme of things...

Sure. There's a major refactoring coming soon in #75. I can release binaries once it's merged shortly.

Hi @kchro3, thanks for reaching out! It needs Conv2D implementation on Metal. I'm currently implementing [LLaVA](https://github.com/haotian-liu/LLaVA/) with llama.cpp and clip.cpp combined (initial release in a few days). The CLIP part...

WIP PR for the initial release of large multimodal model inference is [in another repo](https://github.com/monatis/lmm.cpp/pull/2). Metal implementation of Conv2D will require another pr to the main GGML repo after this...

Yes, I upstreamed clip.cpp to llama.cpp for LLaVA, and now there's some functionality that needs to be downstreamed from llama.cpp back to this repo. Currently I don't have enough time...

@lithdew That would be awesome. And feel free to ping me if you need to ask anything about the implementation.

zsl (zero-shot labeling) examples does that. We need to pass candidate class names with `--text` argument and zsl is scoring those classes.