metadata
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: dpo-selective-buffer-safeipo
results: []
dpo-selective-buffer-safeipo
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3423
- Rewards/chosen: -1.0061
- Rewards/rejected: -1.3040
- Rewards/accuracies: 0.7314
- Rewards/margins: 0.2980
- Rewards/safe Rewards: -0.9953
- Rewards/unsafe Rewards: -1.0113
- Logps/rejected: -222.8744
- Logps/chosen: -231.0451
- Logits/rejected: -0.6287
- Logits/chosen: -1.0992
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-07
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Rewards/safe Rewards | Rewards/unsafe Rewards | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1.0274 | 0.27 | 500 | 0.3731 | -1.0192 | -1.3005 | 0.7075 | 0.2813 | -1.0089 | -1.0281 | -222.5216 | -232.3561 | -0.9855 | -1.4616 |
| 0.9569 | 0.54 | 1000 | 0.3497 | -0.9136 | -1.2026 | 0.7210 | 0.2890 | -0.9006 | -0.9166 | -212.7308 | -221.7959 | -0.5821 | -1.0712 |
| 0.8619 | 0.81 | 1500 | 0.3429 | -0.9136 | -1.1772 | 0.7269 | 0.2635 | -0.9047 | -0.9192 | -210.1883 | -221.8018 | -0.7005 | -1.1466 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.0