nllb-random-custom-tokenizer

This model is a fine-tuned version of Koubra-Gaby/nllb-random-custom-tokenizer on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.0258
• Bleu: 13.4233
• Rouge: 0.3486
• Meteor: 0.2855
• Chrf++: 37.9222

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: 64
• eval_batch_size: 64
• seed: 42
• 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
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Rouge	Meteor	Chrf++
7.9513	1.0331	500	6.2049	0.0651	0.0678	0.0243	3.0229
5.4030	2.0661	1000	4.4498	0.4667	0.1203	0.0733	12.4607
4.0658	3.0992	1500	3.6235	0.5105	0.1266	0.08	13.0516
3.5470	4.1322	2000	3.3902	0.8158	0.1591	0.0935	15.1671
3.3408	5.1653	2500	3.2378	1.1827	0.1702	0.1039	16.5376
3.1927	6.1983	3000	3.1201	1.4013	0.1571	0.1119	18.2268
3.0637	7.2314	3500	3.0126	1.5349	0.1578	0.1131	18.2462
2.9468	8.2645	4000	2.9150	2.0913	0.1801	0.134	19.8481
2.8396	9.2975	4500	2.8305	2.6758	0.192	0.1428	21.2074
2.7448	10.3306	5000	2.7583	3.0791	0.2063	0.1513	22.4461
2.6610	11.3636	5500	2.6929	3.4511	0.2195	0.1595	23.7517
2.5846	12.3967	6000	2.6319	3.6234	0.2232	0.1653	24.2995
2.5092	13.4298	6500	2.5767	4.3742	0.2388	0.1783	25.7717
2.4432	14.4628	7000	2.5312	4.798	0.2466	0.185	26.6679
2.3827	15.4959	7500	2.4874	5.1268	0.2518	0.1907	27.1622
2.3307	16.5289	8000	2.4458	6.1928	0.2674	0.2025	28.6166
2.2697	17.5620	8500	2.4046	6.7166	0.2747	0.2089	29.4449
2.2154	18.5950	9000	2.3663	7.2008	0.2818	0.2179	30.426
2.1653	19.6281	9500	2.3362	7.4984	0.2839	0.2197	30.8209
2.1219	20.6612	10000	2.3028	8.1163	0.2933	0.2261	31.4773
2.0817	21.6942	10500	2.2764	8.4331	0.2969	0.2308	31.9343
2.0388	22.7273	11000	2.2490	8.805	0.3004	0.2349	32.3341
2.0023	23.7603	11500	2.2243	9.0207	0.3011	0.2373	32.7308
1.9710	24.7934	12000	2.2019	9.3553	0.3062	0.2425	33.2971
1.9355	25.8264	12500	2.1801	9.7324	0.3106	0.2466	33.7009
1.9024	26.8595	13000	2.1629	10.1327	0.3142	0.25	34.1051
1.8763	27.8926	13500	2.1436	10.2734	0.3166	0.2538	34.2688
1.8482	28.9256	14000	2.1265	10.6259	0.3201	0.2553	34.7058
1.8234	29.9587	14500	2.1115	10.8501	0.3225	0.2592	35.1492
1.7977	30.9917	15000	2.0983	11.3203	0.3268	0.2631	35.4464
1.7775	32.0248	15500	2.0860	11.5107	0.3285	0.266	35.7986
1.7568	33.0579	16000	2.0761	11.6736	0.3306	0.2696	36.1861
1.7370	34.0909	16500	2.0663	11.8668	0.3329	0.2706	36.3154
1.7189	35.1240	17000	2.0558	12.239	0.3354	0.2735	36.6093
1.7026	36.1570	17500	2.0464	12.2882	0.3367	0.2748	36.7833
1.6889	37.1901	18000	2.0382	12.4823	0.3402	0.2778	37.0442
1.6725	38.2231	18500	2.0273	12.6821	0.3401	0.2803	37.3292
1.6586	39.2562	19000	2.0222	12.848	0.3418	0.2812	37.3796
1.6510	40.2893	19500	2.0162	12.9435	0.3437	0.2835	37.5547
1.6409	41.3223	20000	2.0101	13.1129	0.344	0.2839	37.6632
1.6281	42.3554	20500	2.0054	13.192	0.3443	0.2848	37.8279
1.6179	43.3884	21000	2.0017	13.2852	0.3466	0.2866	37.9176
1.6165	44.4215	21500	1.9983	13.4787	0.3474	0.2872	38.0413
1.6054	45.4545	22000	1.9941	13.4264	0.3478	0.2871	37.9851
1.6019	46.4876	22500	1.9929	13.6088	0.3482	0.2882	38.121
1.5940	47.5207	23000	1.9914	13.6194	0.3492	0.2894	38.188
1.5968	48.5537	23500	1.9898	13.6343	0.3495	0.2894	38.2232
1.5953	49.5868	24000	1.9890	13.66	0.3498	0.2895	38.254
1.5953	50.0	24200	1.9891	13.642	0.3499	0.2896	38.2377

Framework versions

• Transformers 5.18.0
• Pytorch 2.11.0+cu130
• Datasets 4.8.5
• Tokenizers 0.23.2