cil-ordinal-regression-seed2

This model is a fine-tuned version of xlm-roberta-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1651
• Tuned Threshold Mae: 0.3851
• Mae: 0.4468
• Rounded Mae: 0.3860

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: 0.00015
• train_batch_size: 64
• eval_batch_size: 1024
• seed: 2
• 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: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Tuned Threshold Mae	Mae	Rounded Mae
0.3445	0.1411	500	0.2053	0.4647	0.5446	0.4702
0.2047	0.2822	1000	0.1883	0.4296	0.5071	0.4357
0.1921	0.4233	1500	0.1831	0.4184	0.4953	0.4260
0.1912	0.5643	2000	0.1817	0.4161	0.4790	0.4232
0.1856	0.7054	2500	0.1782	0.4075	0.4686	0.4147
0.1806	0.8465	3000	0.1762	0.4040	0.4731	0.4076
0.1777	0.9876	3500	0.1748	0.4008	0.4626	0.4073
0.1701	1.1287	4000	0.1716	0.3966	0.4579	0.3994
0.1713	1.2698	4500	0.1704	0.3925	0.4633	0.3983
0.1718	1.4108	5000	0.1718	0.3913	0.4573	0.4016
0.1674	1.5519	5500	0.1689	0.3908	0.4560	0.3956
0.1665	1.6930	6000	0.1703	0.3890	0.4577	0.3962
0.1668	1.8341	6500	0.1679	0.3887	0.4506	0.3924
0.1658	1.9752	7000	0.1665	0.3878	0.4568	0.3921
0.1606	2.1163	7500	0.1675	0.3867	0.4533	0.3911
0.1581	2.2573	8000	0.1678	0.3856	0.4493	0.3943
0.1591	2.3984	8500	0.1654	0.3855	0.4473	0.3878
0.1581	2.5395	9000	0.1654	0.3854	0.4479	0.3867
0.1575	2.6806	9500	0.1654	0.3855	0.4464	0.3867
0.1571	2.8217	10000	0.1653	0.3849	0.4468	0.3875
0.1587	2.9628	10500	0.1651	0.3851	0.4467	0.3859
0.1587	3.0	10632	0.1651	0.3851	0.4468	0.3860

Framework versions

• PEFT 0.19.1
• Transformers 5.8.1
• Pytorch 2.11.0+cu128
• Datasets 4.8.5
• Tokenizers 0.22.2