metadata
license: apache-2.0
task_categories:
- question-answering
- text-generation
annotations_creators:
- crowdsourced
- expert-generated
language:
- amh
- arb
- ary
- ars
- acq
- arz
- apc
- ben
- ceb
- dan
- deu
- ell
- eng
- eus
- fil
- fin
- fra
- gle
- guj
- hat
- hau
- hin
- hun
- ibo
- ind
- ita
- jav
- jpn
- kan
- kir
- kor
- kur
- lit
- mal
- mar
- mlg
- msa
- mya
- nep
- nld
- nso
- nya
- pan
- pes
- pol
- por
- pus
- rus
- sin
- sna
- snd
- som
- spa
- sqi
- srp
- sun
- swa
- swe
- tam
- tel
- tha
- tur
- ukr
- urd
- vie
- wol
- xho
- yor
- zho
- zul
language_creators:
- crowdsourced
- expert-generated
multilinguality:
- multilingual
size_categories:
- 100K<n<1M
CohereForAI/aya_dataset in ChatML format, ready to use in HuggingFace TRL's SFT Trainer.
Python code used for conversion:
from datasets import load_dataset
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("Felladrin/Llama-160M-Chat-v1")
dataset = load_dataset("CohereForAI/aya_dataset", split="train")
def format(columns):
messages = [
{
"role": "user",
"content": columns["inputs"].strip(),
},
{
"role": "assistant",
"content": columns["targets"].strip(),
},
]
return { "text": tokenizer.apply_chat_template(messages, tokenize=False) }
dataset.map(format).select_columns(['text', 'language', 'language_code', 'annotation_type', 'user_id']).to_parquet("train.parquet")