learning_rate: 2e-05
train_batch_size: 16
eval_batch_size: 32
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
lr_scheduler_warmup_steps: 100
num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Weighted	Accuracy
1.0995	1.0	256	1.0782	0.3230	0.3319	0.4036
1.0417	2.0	512	0.9703	0.5142	0.5198	0.5451
0.9544	3.0	768	1.0096	0.4826	0.4931	0.5270
0.7986	4.0	1024	1.0544	0.5177	0.5216	0.5326
0.6990	5.0	1280	1.0772	0.5387	0.5429	0.5534
0.6277	6.0	1536	1.1651	0.5359	0.5381	0.5437

Framework versions

Transformers 5.0.0
Pytorch 2.10.0+cpu
Datasets 4.8.5
Tokenizers 0.22.2

Metricas de desempeno — 3 configuraciones de batch size

Batch	Epocas	F1-Macro	F1-Weighted	Accuracy	Supera 64%	Mejor
8	6	0.5339	0.54	0.55	NO
16	6	0.5339	0.54	0.55	NO	<- MEJOR
32	6	0.5277	0.53	0.54	NO

El modelo publicado corresponde al batch=16 con F1-Macro=0.5339. XLNet no supero el baseline de BETO (64%) por limitaciones de hardware (float32 obligatorio, sin soporte fp16).

YoliGuadir/xlnet-tass-sentimientos

xlnet-tass-sentimientos

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Metricas de desempeno — 3 configuraciones de batch size

YoliGuadir/
xlnet-tass-sentimientos