I'm working with the meta-llama/Llama-3.2-1B model from Hugging Face Transformers and I only need it to support English. I was wondering if it's possible to remove all the other languages from this model and fine tune it for English-only use cases.

Would doing this result in better memory efficiency and speed during both training and inference? If so, could you provide guidance on how to identify and remove the non-English tokens and language data from the model and tokenizer.

Thanks in advance for your help!