metadata
library_name: peft
tags:
- trl
- dpo
- generated_from_trainer
base_model: Meta-Llama-3-8B-Instruct
model-index:
- name: dpo
results: []
dpo
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.6845
- Rewards/chosen: 0.3768
- Rewards/rejected: 0.3499
- Rewards/accuracies: 0.5373
- Rewards/margins: 0.0269
- Logps/rejected: -233.9739
- Logps/chosen: -254.7565
- Logits/rejected: -0.6114
- Logits/chosen: -0.7300
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-06
- train_batch_size: 1
- eval_batch_size: 1
- seed: 0
- gradient_accumulation_steps: 2
- total_train_batch_size: 2
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 2500
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6689 | 0.0567 | 750 | 0.6852 | 0.2546 | 0.2335 | 0.5373 | 0.0211 | -235.1381 | -255.9783 | -0.6108 | -0.7325 |
| 0.6831 | 0.1133 | 1500 | 0.6835 | 0.3555 | 0.3270 | 0.5597 | 0.0285 | -234.2029 | -254.9690 | -0.6135 | -0.7317 |
| 0.6821 | 0.1700 | 2250 | 0.6855 | 0.3655 | 0.3411 | 0.5485 | 0.0243 | -234.0616 | -254.8697 | -0.6115 | -0.7293 |
Framework versions
- PEFT 0.11.1
- Transformers 4.41.1
- Pytorch 2.3.0+cu121
- Datasets 2.19.1
- Tokenizers 0.19.1