Terjman-Nano-v2.2-512

This model is a fine-tuned version of atlasia/Terjman-Nano on BounharAbdelaziz/Terjman-v2-English-Darija-Dataset-350K. It achieves the following results on atlasia/TerjamaBench set:

• Loss: 3.3996
• Bleu: 18.8386
• Chrf: 38.4027
• Ter: 94.7265
• Gen Len: 9.7059

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: 0.0001
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Ter	Gen Len
13.3826	0.0723	100	3.9345	9.9676	26.2425	104.4045	10.62
12.8584	0.1446	200	3.9364	10.2822	26.9005	118.5236	10.3635
11.6455	0.2169	300	3.9230	11.2322	28.4112	169.8713	11.1788
10.8856	0.2892	400	3.9062	12.1716	29.7281	188.6184	11.4588
9.9054	0.3615	500	3.8526	13.7346	31.3446	173.6084	10.9635
9.1353	0.4338	600	3.7746	14.9908	32.8135	141.8675	10.5882
8.4983	0.5061	700	3.7088	15.7536	34.0031	124.8542	10.0847
7.8704	0.5783	800	3.6546	16.1519	34.6619	125.2093	10.0929
7.5657	0.6506	900	3.6034	16.497	35.4358	124.2517	10.0918
7.1648	0.7229	1000	3.5651	16.8745	36.0108	120.2402	10.0694
6.9344	0.7952	1100	3.5358	17.4455	36.8006	111.3276	9.9271
6.7933	0.8675	1200	3.5146	17.4007	36.7547	109.1005	9.8659
6.5663	0.9398	1300	3.4908	17.8837	37.2366	97.7684	9.6588
6.156	1.0116	1400	3.4978	17.885	37.1388	101.7405	9.68
6.2336	1.0839	1500	3.4947	17.6667	37.0401	104.102	9.7506
6.0831	1.1562	1600	3.4840	17.9693	37.3924	103.7544	9.8318
5.9833	1.2284	1700	3.4736	18.0472	37.6755	100.4721	9.7482
5.889	1.3007	1800	3.4686	18.4178	37.7324	99.5847	9.7494
5.831	1.3730	1900	3.4629	18.546	37.9074	105.798	9.8329
5.7614	1.4453	2000	3.4568	18.4814	38.1828	96.5594	9.6847
5.6739	1.5176	2100	3.4438	18.3851	37.8621	101.9954	9.7859
5.6724	1.5899	2200	3.4434	18.8482	38.0621	100.9195	9.7612
5.5575	1.6622	2300	3.4418	18.4627	38.0205	101.7897	9.8024
5.5368	1.7345	2400	3.4352	18.5974	38.1539	95.8059	9.6659
5.4737	1.8068	2500	3.4317	18.6153	37.9427	95.6594	9.6541
5.5492	1.8791	2600	3.4224	18.7484	38.2565	94.3134	9.64
5.4826	1.9514	2700	3.4228	18.8072	38.2437	101.5818	9.7882
5.3407	2.0231	2800	3.4214	18.572	38.1618	101.9617	9.7929
5.4007	2.0954	2900	3.4154	18.6936	38.1956	100.9953	9.7776
5.3852	2.1677	3000	3.4138	18.8817	38.3229	93.9953	9.6576
5.3565	2.2400	3100	3.4136	18.7169	38.2232	101.2653	9.7824
5.3588	2.3123	3200	3.4117	19.0345	38.5406	93.8038	9.6471
5.3093	2.3846	3300	3.4096	18.8479	38.3863	94.7586	9.6776
5.2726	2.4569	3400	3.4082	18.793	38.4605	93.4033	9.6294
5.3462	2.5292	3500	3.4067	18.7692	38.5134	94.441	9.6506
5.3179	2.6015	3600	3.4082	18.7075	38.3506	95.021	9.6894
5.2932	2.6738	3700	3.4058	18.8893	38.4524	94.1772	9.6635
5.2667	2.7461	3800	3.4101	18.7618	38.4204	95.1496	9.6882
5.2381	2.8184	3900	3.4058	18.7816	38.3548	95.2425	9.6847
5.2377	2.8907	4000	3.4025	18.7729	38.4445	94.5543	9.6812
5.2117	2.9629	4100	3.4036	18.7731	38.6414	94.1615	9.6635
5.1246	3.0347	4200	3.4048	18.8311	38.4772	94.9196	9.6976
5.2024	3.1070	4300	3.4019	18.8853	38.4819	94.7683	9.6988
5.1983	3.1793	4400	3.4031	18.7309	38.3883	95.2313	9.7094
5.2622	3.2516	4500	3.4001	18.7809	38.2732	94.5622	9.6965
5.2596	3.3239	4600	3.4000	18.8192	38.377	94.5234	9.6882
5.2171	3.3962	4700	3.4022	18.8882	38.4257	94.3211	9.6882
5.2709	3.4685	4800	3.4001	18.9978	38.4208	93.8448	9.6706
5.2127	3.5408	4900	3.4010	18.8523	38.4182	94.0805	9.68
5.1972	3.6130	5000	3.4017	18.5275	38.1393	99.7695	9.7859
5.138	3.6853	5100	3.3996	18.624	38.2046	94.6994	9.6929
5.2447	3.7576	5200	3.4008	18.936	38.5847	93.9781	9.6718
5.249	3.8299	5300	3.4007	18.8005	38.3895	94.4262	9.6788
5.1984	3.9022	5400	3.4000	18.7464	38.4657	94.8043	9.6953
5.2089	3.9745	5500	3.4013	18.9012	38.4854	94.7817	9.6882
5.1492	4.0463	5600	3.3997	18.7251	38.3877	94.9903	9.7012
5.1992	4.1186	5700	3.3981	18.7467	38.3938	94.6414	9.6941
5.1869	4.1909	5800	3.3987	18.8237	38.4655	94.1302	9.6765
5.2141	4.2631	5900	3.3992	18.8302	38.3754	100.4665	9.8118
5.1576	4.3354	6000	3.3993	18.6991	38.4918	94.8632	9.7012
5.1984	4.4077	6100	3.4002	18.6786	38.4421	94.547	9.6718
5.1864	4.4800	6200	3.3999	18.845	38.4827	94.5178	9.7
5.1793	4.5523	6300	3.4000	18.6263	38.5282	94.89	9.68
5.2021	4.6246	6400	3.3989	18.621	38.5006	94.5353	9.6765
5.1802	4.6969	6500	3.3991	18.6936	38.361	94.56	9.6906
5.2364	4.7692	6600	3.3991	18.8428	38.6199	94.6669	9.6976
5.1949	4.8415	6700	3.3992	18.7368	38.3819	94.5997	9.6929
5.153	4.9138	6800	3.3994	18.8365	38.5249	94.7585	9.6988
5.2049	4.9861	6900	3.3996	18.8386	38.4027	94.7265	9.7059

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0