M. Yusuf Sarıgöz

If you don't have candidates, CLIP model won't work for image classification. Then I guess your best bet would be to hardcode class names from a common dataset such as...

Yes, or only the classes that you are actually expecting to appear in the image. This is how zero-shot labeling is supposed to work. If you could describe your exact...

Oh got it. This is called image captioning. There exist numerous models for it, but state-of-the-art results come from models like [LLaVA](https://github.com/haotian-liu/LLaVA/). It is basically CLIP + LLaMA bridged with...

@luminosuslight Actually you're right, unfortunately :D I noticed this after training is complete, and that's why the model has difficulty in umlauts sometimes (not always). Anyway, I'm retraining Tacotron2 (and...

It took around 7 secs with the fixed code in #3 in my case (CPU Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz, and TensorFlow 2.5.0-dev20201205). I'll further investigate the performance and...

Hi @DanielFrankee, you have it in [releases](https://github.com/monatis/german-tts/releases).

Hi @lhr0909, thanks for reaching out! Yes I was busy with some contractual work and other higher-priority GGML PRs, mostly for GGUF. Now that it' landed, I'm ready for giving...

I'll implement this in [llama.cpp](https://github.com/ggerganov/llama.cpp/pull/3436), --will be ready this week. It turned out to be easier to start in llama.cpp, then maybe we can move it to this dedicated repo...

Hi @jp555soul you can checkout the logs with the command `docker compose logs`. If it doesn't give a clue for what's going wrong, you can share the request you're trying...

Hi @mdhvg, thanks for raising this PR! I'll review it later today and then merge.