learning_rate: 0.0003
train_batch_size: 4
eval_batch_size: 8
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: linear
lr_scheduler_warmup_steps: 1000.0
training_steps: 10000

Training Loss	Epoch	Step	Validation Loss	Chrf
10.3637	0.4423	1000	3.5018	0.1866
3.8134	0.8846	2000	3.3539	0.0622
3.7061	1.3268	3000	3.3085	0.4353
3.6344	1.7691	4000	3.3038	0.1223
3.6098	2.2114	5000	3.2616	0.3711
3.5840	2.6537	6000	3.2440	0.4423
3.5493	3.0960	7000	3.2275	0.2886
3.5178	3.5383	8000	3.1992	0.1235
3.4959	3.9805	9000	3.1873	0.4834
3.4794	4.4228	10000	3.1865	0.5272

Curiousfox/mT5_tai-lo_to_chinese_LoRA_ver1.0.g

outputs