Marvin
Initial commit
cb0ed1f unverified
metadata
language:
  - de
tags:
  - question-generation
  - german
  - text2text-generation
  - generated_from_trainer
datasets:
  - lmqg/qg_dequad
metrics:
  - bleu4
  - f1
  - rouge
  - exact_match
model-index:
  - name: german-jeopardy-longt5-large-256
    results:
      - task:
          name: Sequence-to-sequence Language Modeling
          type: text2text-generation
        dataset:
          name: lmqg/qg_dequad
          type: default
          args: default
        metrics:
          - name: BLEU-4
            type: bleu4
            value: 4.87
          - name: F1
            type: f1
            value: 23.82
          - name: ROUGE-1
            type: rouge1
            value: 23.88
          - name: ROUGE-2
            type: rouge2
            value: 8.54
          - name: ROUGE-L
            type: rougel
            value: 23.14
          - name: ROUGE-Lsum
            type: rougelsum
            value: 23.13
          - name: Exact Match
            type: exact_match
            value: 0.32

german-jeopardy-longt5-large-256

This model is a fine-tuned version of google/long-t5-tglobal-large on the lmqg/qg_dequad dataset. It achieves the following results on the evaluation set:

  • Loss: 2.8541
  • Brevity Penalty: 0.8795
  • System Length: 18427
  • Reference Length: 20793
  • ROUGE-1: 23.88
  • ROUGE-2: 8.54
  • ROUGE-L: 23.14
  • ROUGE-Lsum: 23.13
  • Exact Match: 0.32
  • BLEU: 4.87
  • F1: 23.82

Model description

See google/long-t5-tglobal-large for more information about the model architecture.
The model was trained on a single NVIDIA RTX 3090 GPU with 24GB of VRAM.

Intended uses & limitations

This model can be used for question generation on German text.

Training and evaluation data

See lmqg/qg_dequad.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0001
  • train_batch_size: 2
  • eval_batch_size: 2
  • seed: 7
  • gradient_accumulation_steps: 128
  • total_train_batch_size: 256
  • optimizer: Adafactor
  • lr_scheduler_type: constant
  • num_epochs: 20

Training results

Training Loss Epoch Step Validation Loss Counts 1 Counts 2 Counts 3 Counts 4 Totals 1 Totals 2 Totals 3 Totals 4 Precisions 1 Precisions 2 Precisions 3 Precisions 4 Brevity Penalty System Length Reference Length ROUGE-1 ROUGE-2 ROUGE-L ROUGE-Lsum Exact Match BLEU Mean Generated Length F1
8.8727 0.99 36 6.3810 2198 0 0 0 2204 0 0 0 99.7278 0.0 0.0 0.0 0.0002 2204 21250 0.0 0.0 0.0 0.0 0.0 0.0 2.0 0.0
6.0165 1.98 72 5.3864 3587 137 0 0 21960 19756 17552 15348 16.3342 0.6935 0.0028 0.0016 1.0 21960 21250 0.0702 0.0079 0.07 0.07 0.0 0.0851 15.0091 0.073
5.1537 3.0 109 4.9617 3601 145 1 0 14449 12245 10041 7837 24.9221 1.1842 0.01 0.0064 0.6246 14449 21250 0.0882 0.0107 0.0877 0.0876 0.0 0.13 9.5309 0.0926
4.863 3.99 145 4.5531 4590 229 19 0 41674 39470 37266 35062 11.0141 0.5802 0.051 0.0014 1.0 41674 21250 0.0811 0.0081 0.0768 0.0767 0.0 0.1468 29.4528 0.0836
4.5201 4.97 181 4.2020 3643 169 19 0 16104 13900 11696 9492 22.6217 1.2158 0.1624 0.0053 0.7265 16104 21250 0.0865 0.0115 0.0856 0.0855 0.0 0.2845 12.5077 0.0907
4.1347 5.99 218 3.9353 3670 167 20 0 16796 14592 12388 10184 21.8504 1.1445 0.1614 0.0049 0.7671 16796 21250 0.087 0.0114 0.0859 0.0858 0.0 0.2878 13.1656 0.0917
4.012 6.98 254 3.7593 3780 198 35 1 16582 14378 12174 9970 22.7958 1.3771 0.2875 0.01 0.7546 16582 21250 0.0916 0.0128 0.0903 0.0902 0.0 0.4139 12.2931 0.0968
3.7048 8.0 291 3.6034 3668 205 36 3 16158 13954 11750 9546 22.7008 1.4691 0.3064 0.0314 0.7297 16158 21250 0.0882 0.0134 0.0873 0.0872 0.0 0.5493 11.7568 0.0923
3.6284 8.99 327 3.4567 4070 527 160 28 17459 15255 13051 10847 23.3118 3.4546 1.226 0.2581 0.8048 17459 21250 0.1109 0.0281 0.1083 0.1082 0.0 1.8083 9.7777 0.1152
3.4605 9.98 363 3.3390 4325 512 128 27 18829 16625 14421 12217 22.9699 3.0797 0.8876 0.221 0.8793 18829 21250 0.1206 0.0288 0.1168 0.1167 0.0 1.6972 12.6729 0.1254
3.2267 10.99 400 3.1995 4498 774 237 49 18802 16598 14394 12190 23.923 4.6632 1.6465 0.402 0.8779 18802 21250 0.1348 0.0405 0.132 0.1319 0.0005 2.5735 11.5009 0.1381
3.1761 11.98 436 3.1165 4578 866 260 50 16963 14759 12555 10351 26.9882 5.8676 2.0709 0.483 0.7767 16963 21250 0.1454 0.0464 0.1426 0.1427 0.0005 2.7554 10.5172 0.1492
3.0323 12.97 472 3.0074 5019 1048 319 59 18077 15873 13669 11465 27.7646 6.6024 2.3337 0.5146 0.839 18077 21250 0.1691 0.0557 0.1648 0.1647 0.0009 3.2318 12.8294 0.1729
2.8223 13.99 509 2.8911 5257 1120 341 85 17074 14870 12666 10462 30.7895 7.5319 2.6922 0.8125 0.783 17074 21250 0.189 0.0635 0.1841 0.184 0.0018 3.7161 12.6824 0.1929
2.7732 14.98 545 2.8103 5616 1271 407 113 17784 15580 13376 11172 31.5789 8.1579 3.0428 1.0115 0.8229 17784 21250 0.2122 0.0731 0.2063 0.2061 0.0045 4.3667 13.0944 0.217
2.58 16.0 582 2.7183 5959 1461 510 171 18808 16604 14400 12196 31.6833 8.7991 3.5417 1.4021 0.8782 18808 21250 0.2286 0.0822 0.2214 0.2212 0.0064 5.357 13.9174 0.2316
2.5368 16.99 618 2.6630 5935 1543 576 201 16923 14719 12515 10311 35.0706 10.483 4.6025 1.9494 0.7744 16923 21250 0.2365 0.089 0.2309 0.2307 0.0059 5.8686 12.3185 0.2377
2.4325 17.98 654 2.5798 6305 1756 685 265 17870 15666 13462 11258 35.2826 11.209 5.0884 2.3539 0.8277 17870 21250 0.2518 0.0982 0.2452 0.2452 0.0059 6.8664 13.1688 0.2537
2.2632 18.99 691 2.5155 6577 1888 762 304 17785 15581 13377 11173 36.9806 12.1173 5.6963 2.7208 0.823 17785 21250 0.2689 0.1102 0.261 0.2611 0.0086 7.5129 13.2373 0.2702
2.2026 19.79 720 2.4997 6644 1853 720 273 17658 15454 13250 11046 37.626 11.9904 5.434 2.4715 0.8159 17658 21250 0.2717 0.1097 0.2628 0.2625 0.0073 7.1987 13.6343 0.2742

Framework versions

  • Transformers 4.32.1
  • Pytorch 2.1.0
  • Datasets 2.12.0
  • Tokenizers 0.13.3