nguni-byt5-large / README.md
francois-meyer's picture
Update README.md
4cc64e4 verified
metadata
license: mit
language:
  - xh
  - nr
  - zu
  - ss

Usage:

  1. Corrupted span prediction.
## Example from here: https://huggingface.co/docs/transformers/en/model_doc/byt5
tokenizer = AutoTokenizer.from_pretrained("francois-meyer/nguni-byt5-large")
model = AutoModelForSeq2SeqLM.from_pretrained("francois-meyer/nguni-byt5-large")
#model = T5ForConditionalGeneration.from_pretrained(model_path)

input_ids_prompt = "The dog chases a ball in the park."
input_ids = tokenizer(input_ids_prompt).input_ids

input_ids = torch.tensor([input_ids[:8] + [258] + input_ids[14:21] + [257] + input_ids[28:]]) ## Corruption

output_ids = model.generate(input_ids, max_length=100)[0].tolist()

output_ids_list = []
start_token = 0
sentinel_token = 258
while sentinel_token in output_ids:
    split_idx = output_ids.index(sentinel_token)
    output_ids_list.append(output_ids[start_token:split_idx])
    start_token = split_idx
    sentinel_token -= 1

output_ids_list.append(output_ids[start_token:])
output_string = tokenizer.batch_decode(output_ids_list)
print(output_string)
  1. For any other task, you will need to fine-tune it like any other T5, mT5, byT5 model.