SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[Flax Generation] Correct inconsistencies PyTorch/Flax (#12662)

Browse Source 
* fix_torch_device_generate_test

* remove @

* correct greedy search

* save intertmed

* add final logits bias

* correct

* up

* add more tests

* fix another bug

* finish tests

* finish marian tests

* up

Co-authored-by: Patrick von Platen <patrick@huggingface.co>
This commit is contained in:
Patrick von Platen
2021-07-13 18:53:30 +01:00
committed by GitHub
parent 7a22a02a70
commit cee2d2135f
7 changed files with 113 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
docs/source/model_doc/marian.rst
View File

@@ -39,25 +39,6 @@ Implementation Notes
- Code to bulk convert models can be found in ``convert_marian_to_pytorch.py``.
- This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__.
Tips
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- In Flax, it is highly advised to pass `early_stopping=True` to `generate`. *E.g.*:
::
>>> from transformers import MarianTokenizer, FlaxMarianMTModel
>>> model = FlaxMarianMTModel.from_pretrained('Helsinki-NLP/opus-mt-en-de')
>>> tokenizer = MarianTokenizer.from_pretrained('Helsinki-NLP/opus-mt-en-de')
>>> text = "My friends are cool but they eat too many carbs."
>>> input_ids = tokenizer(text, max_length=64, return_tensors='jax').input_ids
>>> # Marian has to make use of early_stopping=True
>>> sequences = model.generate(input_ids, early_stopping=True, max_length=64, num_beams=2).sequences
Naming
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~