mbart-large-50-many-to-many-mmt-finetuned-en-to-hi

This model is a fine-tuned version of facebook/mbart-large-50-many-to-many-mmt on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.7540
• Bleu: 9.5451
• Gen Len: 6.3699

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: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
3.2113	0.0	500	2.8535	4.975	6.0562
2.8197	0.01	1000	2.6902	3.2399	6.2449
2.6595	0.01	1500	2.6256	1.2488	6.5596
2.679	0.01	2000	2.5394	4.8497	6.1644
2.5408	0.02	2500	2.4751	2.294	6.3051
2.4689	0.02	3000	2.4052	3.6526	6.3714
2.4299	0.02	3500	2.3642	3.4783	6.4584
2.4303	0.02	4000	2.3545	4.9837	6.343
2.3715	0.03	4500	2.3138	5.3665	6.1981
2.3354	0.03	5000	2.3002	5.8659	6.109
2.3734	0.03	5500	2.2877	3.8468	6.443
2.2968	0.04	6000	2.2385	3.1561	6.4381
2.272	0.04	6500	2.2269	5.7378	6.1155
2.2691	0.04	7000	2.2244	7.8318	6.2325
2.2333	0.05	7500	2.1973	5.4549	6.1463
2.2518	0.05	8000	2.1996	2.3346	6.4954
2.251	0.05	8500	2.1682	4.7228	6.4063
2.252	0.06	9000	2.2211	5.7043	6.3745
2.288	0.06	9500	2.2017	3.7285	6.4893
2.2443	0.06	10000	2.1300	7.0869	6.4448
2.1723	0.06	10500	2.1076	4.3789	6.3482
2.1371	0.07	11000	2.1024	4.2024	6.547
2.1206	0.07	11500	2.1045	7.2662	6.1786
2.1325	0.07	12000	2.0903	3.8746	6.344
2.1315	0.08	12500	2.0817	5.9713	6.4415
2.1285	0.08	13000	2.0637	7.7832	6.1534
2.1223	0.08	13500	2.0522	2.9569	6.313
2.1036	0.09	14000	2.0505	5.0732	6.3308
2.1053	0.09	14500	2.0288	6.2772	6.1638
2.1122	0.09	15000	2.0232	6.8311	6.2005
2.0566	0.1	15500	2.0175	5.8541	6.1907
2.0783	0.1	16000	2.0147	10.0926	6.1531
2.0775	0.1	16500	2.0128	7.6705	6.2624
2.0952	0.1	17000	1.9951	5.5904	6.2104
2.115	0.11	17500	1.9806	8.0092	6.2081
2.0515	0.11	18000	1.9769	5.9444	6.2055
2.0698	0.11	18500	1.9611	8.6585	6.2591
2.0521	0.12	19000	1.9715	9.1678	6.2758
2.0581	0.12	19500	1.9538	7.0038	6.2019
2.0073	0.12	20000	1.9502	7.6102	6.3093
2.0104	0.13	20500	1.9414	7.7584	6.1554
2.0163	0.13	21000	1.9404	5.8758	6.3561
2.03	0.13	21500	1.9294	6.9283	6.1343
1.9915	0.14	22000	1.9159	5.6757	6.2349
2.0158	0.14	22500	1.9234	7.5094	6.1197
1.9616	0.14	23000	1.9170	9.4006	6.2105
1.9954	0.14	23500	1.9008	2.9622	6.2355
2.0116	0.15	24000	1.9026	11.0333	6.0291
1.9742	0.15	24500	1.8973	4.504	6.3386
1.9805	0.15	25000	1.8955	3.8655	6.2335
1.9413	0.16	25500	1.8821	8.9818	6.1769
1.9311	0.16	26000	1.8851	6.7291	6.0846
1.9696	0.16	26500	1.8789	12.1041	6.1274
1.9419	0.17	27000	1.8687	7.2389	6.2407
1.959	0.17	27500	1.8688	7.984	6.3319
1.9449	0.17	28000	1.8655	6.7646	6.2376
1.961	0.18	28500	1.8541	9.8683	6.2369
1.9293	0.18	29000	1.8676	8.2689	6.0779
1.978	0.18	29500	1.8515	5.7599	6.1964
1.9121	0.18	30000	1.8508	7.8691	6.19
1.9566	0.19	30500	1.8350	7.7093	6.1696
1.9279	0.19	31000	1.8455	7.2261	6.2585
1.9717	0.19	31500	1.8374	8.7243	6.2427
1.9215	0.2	32000	1.8239	3.5888	6.242
1.937	0.2	32500	1.8352	3.352	6.2681
1.9103	0.2	33000	1.8260	7.5665	6.2443
1.9363	0.21	33500	1.8122	8.6132	6.1723
1.8938	0.21	34000	1.8175	8.49	6.3157
1.8869	0.21	34500	1.8156	7.8958	6.3069
1.9113	0.22	35000	1.8083	5.658	6.2682
1.9175	0.22	35500	1.8012	9.5439	6.3022
1.9283	0.22	36000	1.8032	9.1064	6.3756
1.9227	0.22	36500	1.7899	8.7293	6.2953
1.9129	0.23	37000	1.7822	4.2586	6.2276
1.8733	0.23	37500	1.7789	9.1095	6.2261
1.8986	0.23	38000	1.7831	4.853	6.2544
1.8655	0.24	38500	1.7762	7.4264	6.3151
1.8996	0.24	39000	1.7648	7.4422	6.3657
1.8771	0.24	39500	1.7690	12.3696	6.1305
1.8719	0.25	40000	1.7565	10.0389	6.2719
1.8993	0.25	40500	1.7481	14.4891	6.2271
1.8756	0.25	41000	1.7516	7.2393	6.2737
1.8195	0.26	41500	1.7534	9.0375	6.2557
1.8384	0.26	42000	1.7540	9.5451	6.3699

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0