João Lages

Hello, I was wondering if anybody has done a Wrapper for SwiftOCR to be able to use it in another language (e.g., Python). Or if there is a way to...

Is dot product the right way to predict?

2

While training implicit sequence models, we use losses like hinge, bpr and pointwise. These losses don't maximize directly the dot product, so why do we use it while predicting?

Could we train Implicit Sequence Models in a multi-label scenario?

4

instead of having only one single label (item) per position, have many instead. I dont know what losses we could use.

Difference between user and sequence representation

6

Can anyone explain me why the user and sequence representation are calculated in [this](https://github.com/maciejkula/spotlight/blob/fa3655996dccdd33b419cef86b4e95da5f1196c0/spotlight/sequence/representations.py#L227) way?? seems like the last state of the LSTM is the sequence representation and the rest...

Is it expected for the training time to not decrease?

2

I'm trying to switch all my embedding and Linear layers by Lora layers. Although the GPU size needed reduces, the training time remains the same, even with less trainable weights....

Thanks for making this repository, really useful :) Let me start with the questions: 1. Is this network specifically done for 28x28 images or would it work for other sizes?...

How did you generate the heatmap plots for the attentions?

2


Can I run the pretrained models on new databases?

3

Is it possible to run the pretrained models on new database+question pairs?

For reference, something like this: https://huggingface.co/hivemind/gpt-j-6B-8bit I'd like to help in this implementation if you're interested in it 😀

Having 2 optimizers

3

Hi there! Thank you for making this implementation open-source! I have one question though: Although you have one backward step, you have 2 optimizers. shouldn't you combine both model's parameters...