Jordi Armengol Estapé

Obtaining unit-wise forward computational graph

1

Hi, thanks for your work. One question: do you know whether it would be feasible to extract the unit-level, forward computational graph of a PyTorch model using some of the...

Hi, thanks for your work. You have tons of Transformer variants implemented in your Github. Just wondering whether you would consider integrating all of them into a single library. Possible...

Hi, many thanks for your work. I'm trying to understand the usage and scope of this library. Can it be used to generate the actual fake definition? Say I have...

[Feature request] Add Fairscale support

3

Adding support for Fairscale would be a great addition.

Change default path of language identifier binary

1

Hi, I think having `os.path.dirname(__file__)` as the default directory for storing the binary would be a better default than `"/tmp/fasttext-langdetect"`, since some systems might not have a working tmp directory....

[Feature Request] "Extensions", a way to separate struct's implementations from their data

2

## Request Disentangle struct's implementations from their data. ### Motivation Improving extensibility (see example below). ### Description and Requirements In the struct definition, define only the attributes: ``` struct MyPair:...

[Feature request] DiVAE

3

DiVAE [1] uses a VQ encoder and a diffusion decoder. Unfortunately, there's no public implementation. It would also be nice to combine that with diffusion Transformers [2]. Any way many...

Load serialized tokenizer with AutoTokenizer

1

How to load a serialized dataset (i.e., defined with a JSON) trained with the tokenizers library using AutoTokenizer? Is it possible? Thanks in advance.