|
--- |
|
tags: |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
model-index: |
|
- name: zephyr-7b-dpo-full |
|
results: [] |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# zephyr-7b-dpo-full |
|
|
|
This model was trained from scratch on the None dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.2027 |
|
- Rewards/chosen: 0.6729 |
|
- Rewards/rejected: -2.3580 |
|
- Rewards/accuracies: 0.9141 |
|
- Rewards/margins: 3.0309 |
|
- Logps/rejected: -380.2658 |
|
- Logps/chosen: -322.0539 |
|
- Logits/rejected: -1.9204 |
|
- Logits/chosen: -1.9591 |
|
|
|
## 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: 8 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 8 |
|
- gradient_accumulation_steps: 2 |
|
- total_train_batch_size: 128 |
|
- total_eval_batch_size: 64 |
|
- 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 | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | |
|
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| |
|
| 0.4898 | 0.12 | 100 | 0.5505 | -0.1967 | -1.0051 | 0.6875 | 0.8085 | -353.2088 | -339.4445 | -1.7659 | -1.8469 | |
|
| 0.4277 | 0.23 | 200 | 0.4655 | -0.4834 | -1.8836 | 0.7383 | 1.4002 | -370.7788 | -345.1795 | -1.7248 | -1.8009 | |
|
| 0.4188 | 0.35 | 300 | 0.3922 | -0.0720 | -2.0263 | 0.7969 | 1.9544 | -373.6328 | -336.9513 | -1.6143 | -1.6899 | |
|
| 0.3506 | 0.46 | 400 | 0.3457 | 0.2171 | -2.0472 | 0.8203 | 2.2643 | -374.0495 | -331.1692 | -1.9794 | -2.0296 | |
|
| 0.3611 | 0.58 | 500 | 0.2959 | 0.2498 | -2.4347 | 0.8516 | 2.6844 | -381.7997 | -330.5164 | -1.8183 | -1.8592 | |
|
| 0.3562 | 0.69 | 600 | 0.2513 | 0.3868 | -2.4732 | 0.8711 | 2.8600 | -382.5696 | -327.7753 | -1.9217 | -1.9736 | |
|
| 0.3624 | 0.81 | 700 | 0.2194 | 0.6454 | -2.3556 | 0.9062 | 3.0010 | -380.2178 | -322.6031 | -1.9301 | -1.9717 | |
|
| 0.4069 | 0.92 | 800 | 0.2027 | 0.6729 | -2.3580 | 0.9141 | 3.0309 | -380.2658 | -322.0539 | -1.9204 | -1.9591 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.38.2 |
|
- Pytorch 2.1.2+cu118 |
|
- Datasets 2.16.1 |
|
- Tokenizers 0.15.2 |
|
|
