metadata
license: apache-2.0
library_name: peft
tags:
- trl
- dpo
- generated_from_trainer
base_model: davidkim205/nox-solar-10.7b-v4
model-index:
- name: dpo_v4_reseve_v1
results: []
dpo_v4_reseve_v1
This model is a fine-tuned version of davidkim205/nox-solar-10.7b-v4 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.2529
- Rewards/chosen: 1.8018
- Rewards/rejected: -2.8948
- Rewards/accuracies: 0.5640
- Rewards/margins: 4.6966
- Logps/rejected: -431.3928
- Logps/chosen: -361.0608
- Logits/rejected: -2.2914
- Logits/chosen: -2.3061
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: 0.0001
- train_batch_size: 2
- eval_batch_size: 2
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1.5803 | 1.0 | 4750 | 1.2529 | 1.8018 | -2.8948 | 0.5640 | 4.6966 | -431.3928 | -361.0608 | -2.2914 | -2.3061 |
Framework versions
- PEFT 0.5.0
- Transformers 4.39.3
- Pytorch 2.1.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2