facebook-NLLB-fr-arb

This model is a fine-tuned version of facebook/nllb-200-distilled-600M on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1409
• Bleu: 27.3675
• Rouge: 0.4694
• Meteor: 0.4058
• Gen Len: 40.2502

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	Gen Len
3.6272	1.0331	500	2.5403	7.5476	0.2713	0.2038	53.6369
2.4300	2.0661	1000	2.0431	12.1379	0.3266	0.2607	47.3629
2.0542	3.0992	1500	1.8090	14.9859	0.3547	0.2903	45.7103
1.8323	4.1322	2000	1.6627	17.1075	0.3726	0.3093	44.2639
1.6809	5.1653	2500	1.5631	18.7848	0.3862	0.3241	43.222
1.5736	6.1983	3000	1.4895	19.6895	0.3943	0.3333	43.0724
1.4786	7.2314	3500	1.4299	20.5418	0.4006	0.3421	42.9126
1.4083	8.2645	4000	1.3886	21.3044	0.4099	0.3495	41.5663
1.3407	9.2975	4500	1.3507	22.1277	0.4155	0.3564	41.7253
1.2851	10.3306	5000	1.3181	22.6505	0.42	0.361	41.9315
1.2374	11.3636	5500	1.2931	23.1023	0.4244	0.3641	41.2145
1.1976	12.3967	6000	1.2733	23.5052	0.4314	0.3707	40.9279
1.1584	13.4298	6500	1.2518	23.9949	0.434	0.3748	41.0879
1.1266	14.4628	7000	1.2369	24.174	0.4363	0.3759	41.023
1.0888	15.4959	7500	1.2221	24.585	0.4403	0.3801	40.9754
1.0633	16.5289	8000	1.2124	24.8795	0.4425	0.3817	40.7984
1.0369	17.5620	8500	1.2021	25.1683	0.4454	0.3856	40.9269
1.0088	18.5950	9000	1.1943	25.4703	0.4485	0.3863	40.763
0.9873	19.6281	9500	1.1860	25.6094	0.4498	0.3885	40.924
0.9688	20.6612	10000	1.1788	25.9172	0.4527	0.3919	40.5743
0.9467	21.6942	10500	1.1698	25.9809	0.4547	0.3933	40.747
0.9288	22.7273	11000	1.1675	26.2421	0.4573	0.3952	40.4807
0.9106	23.7603	11500	1.1621	26.383	0.4561	0.395	40.6878
0.8955	24.7934	12000	1.1596	26.4765	0.4587	0.3975	40.4298
0.8819	25.8264	12500	1.1572	26.7153	0.4623	0.4006	40.3272
0.8685	26.8595	13000	1.1530	26.7297	0.4627	0.4013	40.5053
0.8517	27.8926	13500	1.1468	26.7242	0.4624	0.4005	40.4732
0.8415	28.9256	14000	1.1464	26.9875	0.4641	0.4024	40.4588
0.8295	29.9587	14500	1.1447	27.0239	0.4657	0.4033	40.3425
0.8186	30.9917	15000	1.1432	27.1486	0.4659	0.4041	40.3246
0.8070	32.0248	15500	1.1425	27.3118	0.4669	0.4056	40.3254
0.7993	33.0579	16000	1.1422	27.1232	0.467	0.4048	40.2771
0.7889	34.0909	16500	1.1403	27.2164	0.4675	0.4056	40.4789
0.7810	35.1240	17000	1.1412	27.2592	0.4675	0.4053	40.35
0.7751	36.1570	17500	1.1407	27.3662	0.4687	0.4064	40.3996
0.7672	37.1901	18000	1.1409	27.3675	0.4694	0.4058	40.2502

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2