vik
> I have to downscale images to fit into gradio demo after doing this Interesting, we downscale the image pretty early in the pipeline so I'm not sure what's causing...
Could you share the surrounding code that makes the call? I see a reference to "/home/vizuosense/sensei/engine.py", line 25" in the stack trace. Is there a way I can emulate Raspberry...
Got it, looks like I can emulate with QEMU. Let me try that out.
Understood - working on getting a Raspberry Pi, it's an important platform and I want to make sure it works there.
Update: we now have support for moondream in llama.cpp. Would recommend using that (either directly or via llamafile) when running on Raspberry Pi.
Please feel free to reopen if you run into any issues.
Would absolutely love a community contribution here, I’m currently focused on training the next version so I won’t be able to work on quantizing until after that.
That's right - I had it default to fp32 because it doesn't work on CPU with the current PyTorch implementation. The model itself was trained in 16 bit performance so...
Sounds annoying, what was the prompt? "Say something offensive about the image" worked for me FWIW
If I had to guess it's probably never seen the phrase "roast me" during the training process... the text model (phi-1.5) was trained mostly on synthetic data from ChatGPT so...