metadata
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: dpo-selective-buffer-spo-shift
results: []
dpo-selective-buffer-spo-shift
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 3023.6533
- Rewards/chosen: -0.6906
- Rewards/rejected: -0.6752
- Rewards/accuracies: 0.4671
- Rewards/margins: -0.0155
- Rewards/safe Rewards: -0.6907
- Rewards/unsafe Rewards: -0.6886
- Logps/rejected: -159.8684
- Logps/chosen: -199.9275
- Logits/rejected: -1.4133
- Logits/chosen: -1.7139
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: 2
- gradient_accumulation_steps: 8
- total_train_batch_size: 32
- total_eval_batch_size: 16
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 6785.8539 | 0.27 | 500 | 3119.0669 | -0.6463 | -0.6344 | 0.4765 | -0.0119 | -0.6475 | -0.6439 | -155.7914 | -195.4988 | -1.4656 | -1.7600 |
| 6458.732 | 0.54 | 1000 | 3057.4275 | -0.6406 | -0.6321 | 0.4848 | -0.0085 | -0.6397 | -0.6374 | -155.5639 | -194.9297 | -1.4110 | -1.7036 |
| 5925.3711 | 0.81 | 1500 | 3026.4443 | -0.7053 | -0.6899 | 0.4667 | -0.0154 | -0.7055 | -0.7034 | -161.3376 | -201.3946 | -1.3719 | -1.6829 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.2