Inference

[antirez]

Hi! Thank you for porting MiniGPT. This looks a lot faster than running the PyTorch in Metal, which is very poorly supported indeed (CPU is faster in my tests!). I see that in the model itself generation is implemented, but it is not used anywhere, nor the model is saved after training or tested just to output some text. I wonder if inference would work if I save/reload the model. Thanks.

[vithursant]

hey @antirez thanks! Just a heads up, I am working on a PR for saving/loading pre-trained models for inference. You can expect the feature sometime this week.

[vithursant]

@antirez This issue has been addressed in PR #4. Check it out and let me know, thanks!

[ivanfioravanti]

Model is saved, but inference from it is not working at all, see issue #5