NLP-final-project

This model is a fine-tuned version of FacebookAI/xlm-roberta-large on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1231
• F1 Micro: 0.8344
• F1 Macro: 0.7326

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 0.06
• num_epochs: 40
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Micro	F1 Macro
0.3106	1.0	140	0.2986	0.0	0.0
0.2294	2.0	280	0.2075	0.5667	0.2151
0.1731	3.0	420	0.1717	0.7017	0.4464
0.1240	4.0	560	0.1316	0.7654	0.5373
0.0812	5.0	700	0.1237	0.7710	0.5790
0.0737	6.0	840	0.1214	0.7860	0.6067
0.0599	7.0	980	0.1111	0.8108	0.6502
0.0446	8.0	1120	0.1067	0.8161	0.6664
0.0359	9.0	1260	0.1090	0.8164	0.6773
0.0277	10.0	1400	0.1099	0.8209	0.6826
0.0206	11.0	1540	0.1133	0.8114	0.6649
0.0193	12.0	1680	0.1109	0.8214	0.6812
0.0175	13.0	1820	0.1182	0.8178	0.6996
0.0130	14.0	1960	0.1137	0.8262	0.7195
0.0107	15.0	2100	0.1129	0.8287	0.7033
0.0084	16.0	2240	0.1142	0.8355	0.7179
0.0073	17.0	2380	0.1153	0.8314	0.7213
0.0086	18.0	2520	0.1124	0.8381	0.7254
0.0064	19.0	2660	0.1142	0.8388	0.7269
0.0065	20.0	2800	0.1150	0.8400	0.7367
0.0046	21.0	2940	0.1171	0.8346	0.7174
0.0055	22.0	3080	0.1168	0.8421	0.7333
0.0048	23.0	3220	0.1184	0.8375	0.7252
0.0057	24.0	3360	0.1191	0.8389	0.7425
0.0046	25.0	3500	0.1195	0.8380	0.7264
0.0051	26.0	3640	0.1192	0.8383	0.7407
0.0042	27.0	3780	0.1205	0.8355	0.7315
0.0040	28.0	3920	0.1229	0.8331	0.7263
0.0035	29.0	4060	0.1231	0.8344	0.7326

Framework versions

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