added textbox to use user token, remove text translation for now

fix token acquisition

Improved feedback to the user

fixed button issues

wip

wip

wip

wip

wip

wip

wip

fixed dockerfile to download tikal and build fast_align

wip

add fast align models

Merge branch 'main' of hf.co:spaces/mjuvilla/document-translator

initial commit

Merge pull request #3 from langtech-bsc/any-doc

wip

wip

Merge branch 'main' of hf.co:spaces/mjuvilla/document-translator into any-doc

initial commit

Fixed some bugs, added dockerfile

Removed old code, added support for any* document, disabled "upload" button when we are processing the document

Fixed a lot of error, now the script should crash much less often.

changed nltk tokenizer to multilingual tokenizers

Fixed bug when processing docx files

updated requirements

added language dropdown menus to document translation tab

enforce the usage of the hf token

turns out odts work a bit differently and runs could have more than one tag id, which complicates things quite a lot

Updated readme and added salamandraTA7b translator class

integrated any-doc into the gradle app, separated the translation side to make it easier to implement other translation models

forgot to remove placeholder text

moved scripts to src folder, created new create that hopefully should be able to work with any type of document

Merge pull request #2 from langtech-bsc/multithreading-and-optimizations

added execution time computation for alignments and cleanup old imports

fixed some formatting issues

now both fastaligns run in parallel, also added some improvements to the code (avoid using shell in popen, class Aligner creates the temporary files instead of doing it externally)

Merge pull request #1 from langtech-bsc/windows

removed old file

Added support for windows and linux, removed unused function, added more logs

gradio app for windows

added salamandraTA translation, update requirements

Modified the script so we only run fastalign once instead of once per paragraph, reducing significantly the run time. It involves flattening all the text while keeping the original paragraph index to be able to reconstruct the original structure.

forgot this continue

Fixed issues when dealing with hyperlinks (for now we keep the text and formatting but not the link), also improved format handling and sped things up a bit by avoiding loading fastalign with empty paragraphs

fixed some formatting errors, still haven't fixed line spacing

First commit. For now the translation has not been integrated but reading a docx and writing its translation while keeping the formatting and style should work