BartBanaFinal

This model is a fine-tuned version of IAmSkyDra/BARTBana_v4 on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.9377
• Sacrebleu: 7.6359
• Chrf++: 20.8294

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: 2e-05
• train_batch_size: 256
• eval_batch_size: 256
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 512
• 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
• training_steps: 74268
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Chrf++	Validation Loss	Sacrebleu
No log	0	0	12.4337	43.2284	1.4192
0.0991	0.0269	2000	11.2880	1.2912	1.1187
0.0809	0.0539	4000	11.5611	1.3142	1.3828
0.0753	0.0808	6000	13.3735	1.1897	2.0364
0.0766	0.1077	8000	14.2949	1.1585	2.6931
0.0716	0.1346	10000	15.6571	1.1603	3.1694
0.0683	0.1616	12000	16.1006	1.1283	3.6586
0.0749	0.1885	14000	15.8517	1.1128	3.9157
0.0667	0.2154	16000	16.5723	1.1299	4.4107
0.0667	0.2154	16000	1.1299	4.4107	16.5723
0.0696	0.2424	18000	1.1231	4.8033	16.8438
0.0646	0.2693	20000	1.0649	5.0907	17.1505
0.0561	0.2962	22000	1.1106	5.3579	17.5405
0.0535	0.3232	24000	1.1279	5.6159	17.4801
0.0596	0.3501	26000	1.0895	6.4406	18.6337
0.0586	0.3770	28000	1.0839	6.5442	18.9215
0.0638	0.4039	30000	1.0565	6.4225	18.5386
0.0629	0.4309	32000	1.1079	6.4306	18.5905
0.0632	0.4578	34000	1.0776	6.9070	19.1286
0.0611	0.4847	36000	0.9987	7.2865	19.6978
0.0624	0.5117	38000	1.0437	6.9193	19.0454
0.0664	0.5386	40000	1.0551	6.8607	19.9180
0.0683	0.5655	42000	1.0556	6.8791	19.4370
0.0696	0.5924	44000	0.9795	6.8530	19.4825
0.0731	0.6194	46000	0.9630	6.9622	19.5364
0.0758	0.6463	48000	0.9629	7.4617	20.5333
0.0797	0.6732	50000	0.9573	7.0331	20.3342
0.0735	0.7002	52000	0.9952	6.8602	20.2803
0.0945	0.7271	54000	0.9377	7.6359	20.8294
0.0909	0.7540	56000	0.9200	7.3479	20.3585
0.0937	0.7810	58000	0.8964	7.5754	20.9843
0.102	0.8079	60000	0.9248	7.4648	20.8126
0.1097	0.8348	62000	0.9134	7.4953	21.2279
0.116	0.8617	64000	0.9139	7.1967	20.8760
0.1227	0.8887	66000	0.9069	7.4669	21.0805
0.1426	0.9156	68000	0.8971	7.3416	20.9100
0.1782	0.9425	70000	0.8868	7.3209	20.8953

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1