metadata
license: cc-by-nc-4.0
base_model:
- facebook/nllb-200-distilled-600M
language:
- ami
- trv
- bnn
- pwn
- tay
- tsu
- tao
- dru
- xsy
- pyu
- szy
- ckv
- sxr
- ssf
- xnb
pipeline_tag: translation
datasets:
- ithuan/klokah_asr_train
- ithuan/fb_ilrdf_dict_asr
- ithuan/formosan_db
metrics:
- sacrebleu
model-index:
- name: nllb-600m-formosan-all-finetune
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: ithuan/ithuan_formosan_text
type: ithuan/ithuan_formosan_text
config: Amis_Xiuguluan
split: train
metrics:
- name: ami_Xiug -> zho_Hant (zh)
type: sacrebleu
value: 9.57
- name: zho_Hant -> ami_Xiug (13a)
type: sacrebleu
value: 6.56
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: ithuan/klokah_asr_eval
type: ithuan/klokah_asr_eval
config: Amis_Xiuguluan
split: train
metrics:
- name: ami_Xiug -> zho_Hant (zh)
type: sacrebleu
value: 5.36
- name: zho_Hant -> ami_Xiug (13a)
type: sacrebleu
value: 6.71
Model Card for nllb-600m-formosan-all-finetune
Model Details
nllb-200-distilled-600M finetune on all formosan data (klokah, fb ilrdf dict, formosan_db) without samples only one word.
Training Details
- learning rate: 0.0001
- batch size per gpu: 4
- grad accumulation steps: 1
- epochs: 12
- warmup ratio: 0.1
Uses
please refer https://huggingface.co/docs/transformers/model_doc/nllb