deabuse

Model description

Модель для блюра матерных слов в предложениях, дообученная модель ai-forever/ruT5-base обучалась на искусственном датасете 40к записей. Полноценно текст блюрит не совсем хорошо, однако по отдельности слова вроде неплохо.

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 2e-05
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 4
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 1
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss
5.7858 0.05 400 0.7427
4.3608 0.1 800 0.4047
3.2919 0.15 1200 0.2366
3.2244 0.2 1600 0.2164
3.2491 0.25 2000 0.1757
1.6803 0.3 2400 0.1494
3.2828 0.35 2800 0.1500
3.39 0.4 3200 0.1510
0.0933 0.45 3600 0.1524
3.4757 0.5 4000 0.1423
3.1424 0.55 4400 0.1460
0.9616 0.6 4800 0.1178
2.6271 0.65 5200 0.1178
1.1441 0.7 5600 0.1190
3.018 0.75 6000 0.1136
1.3421 0.8 6400 0.0936
2.3062 0.85 6800 0.0994
2.5594 0.9 7200 0.0945
2.1381 0.95 7600 0.1061
1.0893 1.0 8000 0.1029
0.7525 1.05 8400 0.0978
2.1886 1.1 8800 0.0840
1.9948 1.15 9200 0.0952
0.7933 1.2 9600 0.0871
2.0757 1.25 10000 0.0853
0.6129 1.31 10400 0.0857
0.1338 1.36 10800 0.0936
2.6454 1.41 11200 0.0834
0.4243 1.46 11600 0.0891
0.6615 1.51 12000 0.0885
0.6634 1.56 12400 0.0942
0.5665 1.61 12800 0.0808
0.6661 1.66 13200 0.1021
1.1028 1.71 13600 0.0820
1.5217 1.76 14000 0.0769
0.7644 1.81 14400 0.0771
1.3725 1.86 14800 0.0800
0.846 1.91 15200 0.0788
1.7207 1.96 15600 0.0806
0.9188 2.01 16000 0.0806
1.4303 2.06 16400 0.0814
0.1599 2.11 16800 0.1072
0.1976 2.16 17200 0.0823
0.7077 2.21 17600 0.0830
1.8896 2.26 18000 0.0768
0.6957 2.31 18400 0.0826
0.7827 2.36 18800 0.0802
1.3298 2.41 19200 0.0791
0.2254 2.46 19600 0.0871
1.041 2.51 20000 0.0809
1.5451 2.56 20400 0.0838
1.6318 2.61 20800 0.0801
1.8972 2.66 21200 0.0774
1.8895 2.71 21600 0.0762
0.7721 2.76 22000 0.0740
0.3528 2.81 22400 0.0781
1.325 2.86 22800 0.0770
0.0282 2.91 23200 0.0785
1.6303 2.96 23600 0.0760

Framework versions

  • Transformers 4.39.0.dev0
  • Pytorch 2.2.1
  • Datasets 2.16.1
  • Tokenizers 0.15.2
Downloads last month
15
Safetensors
Model size
223M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for wyluilipe/deabuse

Finetuned
(12)
this model