learning_rate: 1e-05
train_batch_size: 8
eval_batch_size: 16
seed: 1
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: constant_with_warmup
lr_scheduler_warmup_steps: 2000
training_steps: 20000

Training Loss	Epoch	Step	Validation Loss	Chrf
15.9509	0.4742	2000	5.7825	1.0534
6.5120	0.9483	4000	4.5815	1.1035
5.1184	1.4225	6000	4.3163	1.8504
4.9199	1.8966	8000	4.1365	3.1388
4.7272	2.3708	10000	3.9131	3.4953
4.3443	2.8450	12000	3.6709	4.9043
4.1568	3.3191	14000	3.3785	7.7265
3.9503	3.7933	16000	3.1309	9.9968
3.5842	4.2674	18000	2.8839	13.4831
3.4396	4.7416	20000	2.6994	16.3991

Curiousfox/mT5_tai-lo_to_chinese_ver6.1

outputs