-
In #771 I tested the effects of reducing the distillation data to understand that expensive part of our pipeline. However, we should do it again for the `base` student model, as the other one was done…
-
### Feature request
It look like ONNX now supports 4bit: https://onnx.ai/onnx/technical/int4.html
It would be nice if we could use 4bit models with transformers.js .
### Motivation
Make models f…
-
Navigate to https://github.com/facebookresearch/fairseq/tree/nllb and clone the repo. Before running the install instructions, add the following to the setup.py script:
After line 270, add the foll…
-
### Model description
With the recent support for custom models, is it possible to run [IndicTrans2](https://huggingface.co/ai4bharat/indictrans2-en-indic-dist-200M) . It is basically nllb with a cus…
-
As the second recipe after NLLB, write the w2v-BERT (and wav2vec2) pretraining recipe for users to check out. This will likely branch to several subtasks once we start working on it.
-
There are 275.5 million population in Indonesia. Can you support Bahasa Indonesia?
-
#### What is your question?
How to use the language identification model trained on Flores-200 (mentioned in the NLLB paper) ? Model is presented in the repo but the utility of the LID model via co…
-
Each of the micro service in VE could maintain a JSON file with the items whose versions it want to expose in its code base. And an API could be used to fetch these values.
Example, In CMS, as we h…
-
Hi,
I have used all the different NLLB models for Japanese to English, and English to Japanese translations. I have observed that the translation quality of NLLB-200(Dense, 3.3B) is very bad when c…
-
I'm using the below code which will try to translate from Romanian to English
```
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("facebo…