Kerim Buyukakyuz

Then why the [doc](https://huggingface.co/docs/transformers/en/model_doc/nllb#generating-with-nllb) says otherwise? This is `V4.42.0`. I also don't understand how to use `tokenizer.encode("deu_Latn")[0]`. What's the keyword? Is this a positional argument? @ArthurZucker

This is just OLoRA but starting from random weights. How can starting from random weights, rather than getting that information from pretrained weights, converge faster? Did you actually run tests?...

I honestly don't see the performance benefit. But if you think there is an ease of use benefit, there could be some value. This goes for every other decomposition method,...

Weird. We don't do anything model specific, we rely on PEFT's quantization. It should be related to that.