IAmSkyDra/BartBanaFinal_v3

BartBanaFinal

This model is a fine-tuned version of IAmSkyDra/BARTBana_v5 on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.9120
• Sacrebleu: 5.0981
• Chrf++: 18.4403

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: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Chrf++	Validation Loss	Sacrebleu
No log	0	0	11.1997	74.5460	0.9579
0.1371	0.0269	2000	7.5555	1.4283	0.1498
0.1246	0.0539	4000	8.2284	1.4896	0.3342
0.1192	0.0808	6000	10.0258	1.4069	0.5165
0.1232	0.1077	8000	9.0570	1.2832	0.4636
0.1281	0.1346	10000	9.6671	1.1653	0.8182
0.1166	0.1616	12000	11.0435	1.2052	0.8000
0.1365	0.1885	14000	11.6117	1.2694	1.0365
0.1369	0.2154	16000	11.8559	1.1566	1.5111
0.1435	0.2424	18000	10.1954	1.1604	1.1035
0.1309	0.2693	20000	12.1724	1.1573	1.8697
0.1429	0.2962	22000	11.8521	1.1820	1.3905
0.1479	0.3232	24000	12.8547	1.0645	1.9644
0.1033	0.3501	26000	12.3747	1.1773	1.7998
0.1068	0.3770	28000	11.6427	1.2409	1.3029
0.1139	0.4039	30000	13.7875	1.1162	2.3469
0.1139	0.4039	30000	13.7875	1.1162	2.3469
0.1287	0.4309	32000	12.3235	1.1047	1.5293
0.1369	0.4578	34000	12.0428	1.1774	1.7930
0.1338	0.4847	36000	12.5997	1.1800	1.8025
0.1398	0.5117	38000	13.9852	1.1218	2.1374
0.1793	0.5386	40000	14.9792	1.0166	2.8086
0.1556	0.5655	42000	14.8273	1.0475	2.8293
0.1563	0.5924	44000	14.8104	1.0428	2.5901
0.1373	0.6194	46000	14.0005	1.0972	2.3887
0.1686	0.6463	48000	14.6327	1.0907	2.5002
0.2061	0.6732	50000	17.3738	0.9981	3.1007
0.175	0.7002	52000	17.6129	0.9794	4.3472
0.2017	0.7271	54000	15.4695	0.9891	3.2719
0.1949	0.7540	56000	16.5992	0.9531	4.4438
0.2277	0.7810	58000	18.9529	0.9621	4.6969
0.2273	0.8079	60000	17.1567	0.9442	4.5503
0.2572	0.8348	62000	17.4794	0.9348	4.5176
0.2874	0.8617	64000	17.2450	0.9010	4.5275
0.2548	0.8887	66000	18.4403	0.9120	5.0981
0.3086	0.9156	68000	16.9905	0.9026	4.4480
0.4937	0.9425	70000	16.6739	0.8929	4.3731
0.6425	0.9695	72000	17.5479	0.8724	4.6322
0.8591	0.9964	74000	15.8401	0.8449	4.6105
0.9277	1.0	74268	0.8437	4.6162	15.9720

Framework versions

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