ContriBERT-ACL

This model is a fine-tuned version of allenai/scibert_scivocab_uncased on taln-ls2n/ARRContributions. It achieves the following results on the evaluation sets:

Evaluation Set	F1 Micro	F1 Macro	Loss
Author-Annotated	0.5804	0.3775	0.3232
Expert-Annotated	0.6542	0.3982	0.2682

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: 1e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 0
• 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
• num_epochs: 20
• early_stopping_patience: 5

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Micro	F1 Macro
0.4687	1.0	51	0.3894	0.3497	0.0617
0.3926	2.0	102	0.3627	0.4771	0.1660
0.3624	3.0	153	0.3412	0.5007	0.1739
0.3444	4.0	204	0.3302	0.5205	0.1841
0.328	5.0	255	0.3234	0.5365	0.2173
0.3127	6.0	306	0.3196	0.5447	0.2463
0.2989	7.0	357	0.3244	0.5457	0.2639
0.2848	8.0	408	0.3177	0.5596	0.3584
0.2719	9.0	459	0.3171	0.5688	0.3540
0.2627	10.0	510	0.3192	0.5763	0.3616
0.2502	11.0	561	0.3194	0.5818	0.3868
0.2405	12.0	612	0.3246	0.5788	0.3713
0.2337	13.0	663	0.3195	0.5844	0.3928
0.227	14.0	714	0.3262	0.5696	0.3898
0.2192	15.0	765	0.3227	0.5796	0.3792
0.2145	16.0	816	0.3250	0.5751	0.3779
0.2096	17.0	867	0.3256	0.5773	0.3776
0.2074	18.0	918	0.3277	0.5740	0.3792

Framework versions

• Transformers 4.57.1
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.22.1