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: 18010.9043
- Rewards/chosen: -1.5473
- Rewards/rejected: -1.8909
- Rewards/accuracies: 0.7429
- Rewards/margins: 0.3436
- Rewards/safe Rewards: -1.5421
- Rewards/unsafe Rewards: -1.5499
- Logps/rejected: -281.5577
- Logps/chosen: -285.1670
- Logits/rejected: 3.9983
- Logits/chosen: 2.5248
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 48290.35 | 0.27 | 500 | 18207.4277 | -0.8680 | -1.2092 | 0.7308 | 0.3412 | -0.8648 | -0.8682 | -213.3933 | -217.2423 | 0.5180 | -0.2676 |
| 45912.9125 | 0.54 | 1000 | 18076.0898 | -1.4658 | -1.8259 | 0.7394 | 0.3601 | -1.4604 | -1.4691 | -275.0630 | -277.0186 | 3.7605 | 2.3434 |
| 42808.3344 | 0.81 | 1500 | 18052.3926 | -1.6663 | -1.9853 | 0.7414 | 0.3191 | -1.6613 | -1.6696 | -291.0034 | -297.0653 | 4.3253 | 2.8214 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.0