yesj1234/koja_mbartLarge_37p_exp2

mbartLarge_koja_37p_exp2

This model is a fine-tuned version of facebook/mbart-large-50-many-to-many-mmt on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8988
• Bleu: 6.7577
• Gen Len: 17.8104

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 350
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
2.0622	0.11	1250	1.6679	1.2834	17.8009
1.5139	0.22	2500	1.4378	2.0427	17.8496
1.4121	0.33	3750	1.3116	2.7599	17.7667
1.2879	0.44	5000	1.2381	3.1444	17.8887
1.2344	0.55	6250	1.1769	3.3835	17.8323
1.1778	0.66	7500	1.1382	3.9511	17.4892
1.1461	0.77	8750	1.0938	3.9402	18.0136
1.1151	0.88	10000	1.0749	4.2134	18.0537
1.093	0.99	11250	1.0418	3.9587	17.8715
1.0626	1.1	12500	1.0315	4.6251	17.9406
1.0192	1.21	13750	1.0132	4.9573	18.1266
0.9957	1.32	15000	0.9989	4.3068	18.0925
0.9778	1.43	16250	0.9850	5.0517	17.8783
0.9446	1.54	17500	0.9748	5.0194	17.9348
0.9236	1.65	18750	0.9619	4.6011	17.7926
0.9091	1.76	20000	0.9564	4.6035	17.9399
0.9072	1.87	21250	0.9533	4.8313	17.6221
0.8758	1.98	22500	0.9421	5.2707	17.5851
0.8539	2.09	23750	0.9304	5.2661	17.821
0.8575	2.2	25000	0.9329	4.9143	17.8879
0.8314	2.31	26250	0.9262	5.106	18.0037
0.8248	2.42	27500	0.9241	5.3073	17.6632
0.8151	2.53	28750	0.9302	5.5675	17.7676
0.8093	2.64	30000	0.9149	6.2644	17.8475
0.7691	2.75	31250	0.8988	6.6682	17.7685
0.771	2.86	32500	0.9189	5.7856	17.8678
0.7658	2.97	33750	0.9175	6.2468	17.7313
0.7914	3.08	35000	0.9020	5.5525	17.7627
0.7264	3.19	36250	0.9046	6.2055	17.7662

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1