en-vi-model_v3_opus

This model is a fine-tuned version of t5-small on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8572
• Bleu: 8.2434

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Bleu	Validation Loss
1.4493	0.13	500	3.9998	1.3541
1.2915	0.26	1000	5.3936	1.2113
1.2059	0.38	1500	5.8381	1.1367
1.1573	0.51	2000	6.2422	1.0901
1.1121	0.64	2500	6.6271	1.0542
1.0867	0.77	3000	6.8796	1.0252
1.0623	0.9	3500	7.0393	1.0068
1.0408	1.02	4000	7.2660	0.9882
1.0203	1.15	4500	7.0553	0.9723
1.0054	1.28	5000	7.4555	0.9624
0.9977	1.41	5500	7.4260	0.9526
0.9931	1.54	6000	7.5231	0.9396
0.9804	1.66	6500	7.4376	0.9324
0.9691	1.79	7000	7.5227	0.9264
0.9645	1.92	7500	7.6859	0.9193
0.9509	2.05	8000	7.6473	0.9144
0.9485	2.18	8500	7.6548	0.9118
0.9437	2.3	9000	7.6066	0.9073
0.9393	2.43	9500	7.7140	0.9019
0.9336	2.56	10000	7.8095	0.8970
0.9368	2.69	10500	7.9377	0.8937
0.925	2.82	11000	7.8425	0.8898
0.921	2.94	11500	7.9008	0.8864
0.9177	3.07	12000	7.9134	0.8836
0.9151	3.2	12500	0.8821	7.8647
0.9104	3.33	13000	0.8790	8.0830
0.9035	3.46	13500	0.8766	8.0959
0.8992	3.58	14000	0.8741	8.0178
0.8986	3.71	14500	0.8720	8.0384
0.894	3.84	15000	0.8683	8.0913
0.8932	3.97	15500	0.8663	8.0997
0.8889	4.1	16000	0.8641	8.1088
0.8888	4.22	16500	0.8629	8.0665
0.8856	4.35	17000	0.8607	8.2836
0.8826	4.48	17500	0.8613	8.2354
0.8862	4.61	18000	0.8578	8.1166
0.8811	4.74	18500	0.8583	8.1473
0.8799	4.86	19000	0.8579	8.1836
0.8827	4.99	19500	0.8572	8.2434

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu118
• Datasets 2.14.5
• Tokenizers 0.14.1