Edit model card

amazon-reviews-input-output-1.3b

This model is a fine-tuned version of facebook/opt-1.3b on the AlekseyKorshuk/amazon-reviews-input-output dataset. It achieves the following results on the evaluation set:

  • Loss: 3.5488
  • Accuracy: 0.0355
  • Samples: 100
  • Perplexity: 34.7725
  • Table: <wandb.data_types.Table object at 0x7ffa3c3fd700>

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: 5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • total_train_batch_size: 64
  • total_eval_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 5.0

Training results

Training Loss Epoch Step Validation Loss Accuracy
3.2024 0.06 1 2.9121 0.0385
3.1226 0.12 2 2.9121 0.0385
3.1321 0.19 3 2.8477 0.0394
2.9875 0.25 4 2.8477 0.0394
2.9717 0.31 5 2.8555 0.0391
2.9341 0.38 6 2.8438 0.0392
3.0376 0.44 7 2.8184 0.0396
2.8164 0.5 8 2.7988 0.0395
3.0857 0.56 9 2.7988 0.0394
2.9492 0.62 10 2.7969 0.0395
2.8633 0.69 11 2.7969 0.0395
2.8994 0.75 12 2.7910 0.0398
3.0024 0.81 13 2.7812 0.0401
2.937 0.88 14 2.7812 0.0399
2.9963 0.94 15 2.7812 0.0399
3.0168 1.0 16 2.7754 0.04
2.2589 1.06 17 2.7715 0.0397
2.2568 1.12 18 2.7793 0.0395
2.3138 1.19 19 2.8027 0.0393
2.2759 1.25 20 2.8184 0.0393
2.5137 1.31 21 2.8262 0.0390
2.2997 1.38 22 2.8320 0.0388
2.2693 1.44 23 2.8359 0.0392
2.204 1.5 24 2.8379 0.0387
2.3713 1.56 25 2.8359 0.0391
2.3448 1.62 26 2.8340 0.0391
2.217 1.69 27 2.8359 0.0391
2.3082 1.75 28 2.8379 0.0385
2.2878 1.81 29 2.8379 0.0386
2.2429 1.88 30 2.8379 0.0385
2.2838 1.94 31 2.8359 0.0385
2.4038 2.0 32 2.8379 0.0387
1.8481 2.06 33 2.8555 0.0384
1.657 2.12 34 2.8965 0.0382
1.6996 2.19 35 2.9590 0.0380
1.6741 2.25 36 3.0312 0.0379
1.594 2.31 37 3.0410 0.0380
1.5201 2.38 38 3.0156 0.0381
1.5149 2.44 39 3.0137 0.0380
1.5521 2.5 40 3.0176 0.0379
1.5364 2.56 41 3.0273 0.0378
1.5385 2.62 42 3.0391 0.0380
1.4794 2.69 43 3.0488 0.0380
1.4313 2.75 44 3.0527 0.0378
1.5071 2.81 45 3.0469 0.0378
1.4799 2.88 46 3.0449 0.0378
1.521 2.94 47 3.0371 0.0380
1.4603 3.0 48 3.0410 0.0379
1.25 3.06 49 3.0859 0.0381
1.0411 3.12 50 3.1797 0.0375
1.0385 3.19 51 3.2969 0.0371
1.0254 3.25 52 3.3613 0.0367
0.9656 3.31 53 3.3633 0.0368
1.036 3.38 54 3.3359 0.0366
0.9366 3.44 55 3.2949 0.0366
0.9712 3.5 56 3.2695 0.0367
1.0066 3.56 57 3.2676 0.0366
0.9952 3.62 58 3.2773 0.0368
1.0352 3.69 59 3.2891 0.0367
1.0212 3.75 60 3.3164 0.0362
0.9468 3.81 61 3.3203 0.0360
0.9155 3.88 62 3.3223 0.0366
0.8552 3.94 63 3.3262 0.0370
0.9575 4.0 64 3.3340 0.0370
0.6384 4.06 65 3.375 0.0370
0.6436 4.12 66 3.4453 0.0364
0.5752 4.19 67 3.5391 0.0358
0.6542 4.25 68 3.6016 0.0354
0.6724 4.31 69 3.6016 0.0354
0.591 4.38 70 3.5938 0.0359
0.5346 4.44 71 3.5801 0.0361
0.5112 4.5 72 3.5762 0.0361
0.5443 4.56 73 3.5840 0.0362
0.5689 4.62 74 3.6152 0.0358
0.5667 4.69 75 3.6328 0.0358
0.554 4.75 76 3.6348 0.0357
0.6087 4.81 77 3.625 0.0355
0.5236 4.88 78 3.6152 0.0355
0.5458 4.94 79 3.5781 0.0355
0.5702 5.0 80 3.5488 0.0355

Framework versions

  • Transformers 4.25.0.dev0
  • Pytorch 1.12.1+cu113
  • Datasets 2.3.2
  • Tokenizers 0.12.1
Downloads last month
10
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 AlekseyKorshuk/amazon-reviews-input-output-1.3b

Evaluation results

  • Accuracy on AlekseyKorshuk/amazon-reviews-input-output
    self-reported
    0.036