OpenAI's Whisper model is a really great one, supports many languages ans translationa and is available both as a pretrained model (https://github.com/openai/whisper) which can be selfhosted and as an API (https://openai.com/pricing).
I just found out this was foss yesterday, and really want to start playing with it on my system. Is it GPU-agnostic? I have an AMD card and don't see any mention of GPU support or CUDA requirement on the github docs.
AMD has ROCm which is available on And Radeon Instinct GPUs (server GPUs) and some consumer GPUs. You'd need to double check whether your GPU supports ROCm.
It seems there is some discussion happening here on the use of ROCm with Whisper: https://github.com/openai/whisper/discussions/105 And here (showing it might be possible?): https://github.com/openai/whisper/discussions/55
I also found this which could be of interest:
MLC-LLM, which "Enable everyone to develop, optimize and deploy AI models natively on everyone's devices."
Here used to deploy Llama-2-13B on the RX 7900 XTX:
https://blog.mlc.ai/2023/08/09/Making-AMD-GPUs-competitive-for-LLM-inference?ref=upstract.com
Thanks for that, I've been able to get Stable Diffusion running locally with ROCm so it looks like it should be possible then.
Not great yet and not even good for my accent, but dicio is the only foss one I've got working on android so far.
If it's free and open source and it's also software, it can be discussed here. Subcommunity of Technology.
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.