metadata
library_name: transformers
license: mit
base_model: openai-community/gpt2
tags:
- trl
- dpo
- generated_from_trainer
datasets:
- piqa
model-index:
- name: dpo-model-output
results: []
dpo-model-output
This model is a fine-tuned version of openai-community/gpt2 on the piqa dataset. It achieves the following results on the evaluation set:
- Loss: 1.0022
- Rewards/chosen: -17.7008
- Rewards/rejected: -19.2025
- Rewards/accuracies: 0.6638
- Rewards/margins: 1.5016
- Logps/rejected: -283.3009
- Logps/chosen: -266.5706
- Logits/rejected: -79.0250
- Logits/chosen: -79.1854
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: 4
- eval_batch_size: 4
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.717 | 1.0 | 3223 | 0.6773 | -6.7817 | -7.2423 | 0.6513 | 0.4606 | -163.6990 | -157.3790 | -76.4738 | -76.4513 |
| 0.3122 | 2.0 | 6446 | 0.7910 | -12.1665 | -13.1665 | 0.6675 | 1.0000 | -222.9408 | -211.2272 | -85.8639 | -85.9837 |
| 0.0481 | 3.0 | 9669 | 1.0022 | -17.7008 | -19.2025 | 0.6638 | 1.5016 | -283.3009 | -266.5706 | -79.0250 | -79.1854 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.0.0
- Datasets 2.16.1
- Tokenizers 0.19.1