Edit model card

pos-polish-gpt2-small

This model was trained from polish-gpt2-small on clarin-pl/nkjp-pos dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3109
  • Precision: 0.8793
  • Recall: 0.9255
  • F1: 0.9018
  • Accuracy: 0.9371

Model description

Trained from polish-gpt2-small

Intended uses & limitations

Part-of-speech tagging for Polish language. Tags description at the bottom of http://nkjp.pl/poliqarp/help/plse2.html

Training and evaluation data

Dataset: clarin-pl/nkjp-pos

Datacollator:

from transformers import DataCollatorForTokenClassification
data_collator = DataCollatorForTokenClassification(tokenizer=tokenizer)

Training procedure

GPU: RTX 3090

Training time: 00:50:24

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 2e-05
  • train_batch_size: 16
  • eval_batch_size: 2
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 10
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Precision Recall F1 Accuracy
0.0 0 3.6116 0.0464 0.0524 0.0492 0.0676
0.2303 1.0 1222 0.2159 0.8737 0.9225 0.8974 0.9347
0.1776 2.0 2444 0.2124 0.8799 0.9254 0.9021 0.9381
0.1467 3.0 3666 0.2205 0.8759 0.9241 0.8994 0.9368
0.1254 4.0 4889 0.2304 0.8792 0.9256 0.9018 0.9377
0.1091 5.0 6111 0.2480 0.8787 0.9251 0.9013 0.9375
0.0949 6.0 7333 0.2651 0.8794 0.9250 0.9016 0.9373
0.0857 7.0 8555 0.2794 0.8791 0.9251 0.9015 0.9372
0.079 8.0 9778 0.2922 0.8789 0.9247 0.9012 0.9366
0.0736 9.0 11000 0.3037 0.8807 0.9256 0.9026 0.9375
0.0691 10.0 12220 0.3109 0.8793 0.9255 0.9018 0.9371

Framework versions

  • Transformers 4.36.2
  • Pytorch 2.1.2+cu121
  • Datasets 2.16.1
  • Tokenizers 0.15.0
Downloads last month
8
Safetensors
Model size
126M 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.

Dataset used to train nie3e/pos-polish-gpt2-small