metadata
license: llama3.1
library_name: peft
tags:
- trl
- dpo
- generated_from_trainer
base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
model-index:
- name: llama3.1_8b_dpo_bwgenerator_test2
results: []
llama3.1_8b_dpo_bwgenerator_test2
This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0977
- Rewards/chosen: -12.6599
- Rewards/rejected: -40.7969
- Rewards/accuracies: 0.9925
- Rewards/margins: 28.1371
- Logps/rejected: -518.0425
- Logps/chosen: -211.2417
- Logits/rejected: -1.2740
- Logits/chosen: -1.8549
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: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.065 | 0.2396 | 1000 | 0.1978 | -24.9705 | -63.8717 | 0.9881 | 38.9011 | -748.7902 | -334.3486 | -1.2031 | -1.8039 |
| 0.0803 | 0.4793 | 2000 | 0.1339 | -16.1506 | -46.8097 | 0.9925 | 30.6591 | -578.1700 | -246.1489 | -1.2214 | -1.8149 |
| 0.0588 | 0.7189 | 3000 | 0.1012 | -12.8597 | -41.3756 | 0.9925 | 28.5159 | -523.8289 | -213.2401 | -1.2775 | -1.8541 |
| 0.0422 | 0.9585 | 4000 | 0.0977 | -12.6599 | -40.7969 | 0.9925 | 28.1371 | -518.0425 | -211.2417 | -1.2740 | -1.8549 |
Framework versions
- PEFT 0.10.0
- Transformers 4.44.0
- Pytorch 2.3.0+cu121
- Datasets 2.14.7
- Tokenizers 0.19.1