Fix for nllb

[Galoist]

Fixes #63

[xhluca]

Will merge once the tests pass!

[xhluca]

Would it be possible to change the minimum version for this in the setup.py, so users with older versions before the function was added can upgrade their version of transformers?

[xhluca]

Right now it is transformers>=4.30.2 but i'm uncertain if that feature existed in 4.30

[Galoist]

It seems to work with transformers==4.30.2

pixi.toml:

[project]
authors = ["..."]
channels = ["conda-forge"]
description = "Add a short description here"
name = "translate"
platforms = ["osx-arm64"]
version = "0.1.0"

[tasks]

[dependencies]
python = "==3.11"
ipython = ">=8.27.0,<9"

[pypi-dependencies]
dl-translate = { git = "https://github.com/Galoist/dl-translate" }
transformers = "==4.30.2"

In [1]: import dl_translate as dlt
   ...:

In [2]: translate_model: str = "facebook/nllb-200-distilled-600M"

In [3]:         mt = dlt.TranslationModel(translate_model, device="mps")
/private/tmp/translate/.pixi/envs/default/lib/python3.11/site-packages/huggingface_hub/file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(
/private/tmp/translate/.pixi/envs/default/lib/python3.11/site-packages/huggingface_hub/file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(
/private/tmp/translate/.pixi/envs/default/lib/python3.11/site-packages/transformers/modeling_utils.py:463: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  return torch.load(checkpoint_file, map_location="cpu")

In [4]: mt.translate("Привіт, світ!", dlt.lang.nllb200.UKRAINIAN, dlt.lang.nllb200.FRENCH)
Out[4]: 'Je vous salue, mon ami.'

[xhluca]

Awesome! In that case, i'll merge now