---
language: no
license: CC-BY 4.0
tags:
- seq2seq
datasets:
- Norwegian Nynorsk/Bokmål

---
# 🇳🇴 Norwegian T5 Base model Trained on the NCC🇳🇴  

This is a Norwegian T5-base model trained on the Norwegian Colossal Corpus (NCC) on a TPU v3-8. It needs to be finetuned on a specific task before being used for anything.


 The following setting were used in training:
```bash
./run_t5_mlm_flax_streaming.py \\n    --output_dir="./" \\n    --model_type="t5" \\n    --config_name="./" \\n    --tokenizer_name="./" \\n    --dataset_name="pere/norwegian_colossal_corpus_v2_short100k" \\n    --max_seq_length="512" \\n    --weight_decay="0.01" \\n    --per_device_train_batch_size="32" \\n    --per_device_eval_batch_size="32" \\n    --learning_rate="8e-3" \\n    --warmup_steps="5000" \\n    --overwrite_output_dir \\n    --cache_dir /mnt/disks/flaxdisk/cache/ \\n    --num_train_epochs="5" \\n    --adam_beta1="0.9" \\n    --adam_beta2="0.98" \\n    --logging_steps="500" \\n    --num_train_steps="1000000" \\n    --num_eval_samples="5000" \\n    --save_steps="5000" \\n    --eval_steps="5000" \\n    --preprocessing_num_workers 96 \\n    --adafactor \\n    --push_to_hub
 ```