metadata
license: apache-2.0
library_name: peft
tags:
- alignment-handbook
- generated_from_trainer
- trl
- dpo
- generated_from_trainer
datasets:
- snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset
base_model: mistralai/Mistral-7B-Instruct-v0.2
model-index:
- name: zephyr-7b-dpo-qlora-pairrm
results: []
zephyr-7b-dpo-qlora-pairrm
This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset dataset. It achieves the following results on the evaluation set:
- Loss: 0.6773
- Rewards/chosen: -1.5442
- Rewards/rejected: -1.6837
- Rewards/accuracies: 0.5687
- Rewards/margins: 0.1395
- Logps/rejected: -394.7031
- Logps/chosen: -375.1367
- Logits/rejected: -4.4436
- Logits/chosen: -4.4568
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: 4
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6909 | 0.08 | 100 | 0.6921 | -0.0169 | -0.0192 | 0.5427 | 0.0023 | -228.2522 | -222.4038 | -2.6219 | -2.6247 |
| 0.684 | 0.16 | 200 | 0.6873 | -0.0803 | -0.0944 | 0.5567 | 0.0141 | -235.7721 | -228.7468 | -2.7599 | -2.7629 |
| 0.6795 | 0.24 | 300 | 0.6839 | -0.5138 | -0.5516 | 0.5460 | 0.0378 | -281.4856 | -272.0929 | -3.5141 | -3.5199 |
| 0.6561 | 0.32 | 400 | 0.6812 | -0.8158 | -0.8788 | 0.5573 | 0.0630 | -314.2105 | -302.2954 | -3.7484 | -3.7580 |
| 0.633 | 0.4 | 500 | 0.6787 | -0.9027 | -0.9810 | 0.5597 | 0.0782 | -324.4269 | -310.9858 | -4.0978 | -4.1077 |
| 0.6302 | 0.48 | 600 | 0.6785 | -1.1692 | -1.2692 | 0.5597 | 0.1000 | -353.2493 | -337.6355 | -4.4318 | -4.4435 |
| 0.5743 | 0.56 | 700 | 0.6835 | -1.5435 | -1.6640 | 0.5630 | 0.1205 | -392.7273 | -375.0575 | -4.5047 | -4.5182 |
| 0.6443 | 0.64 | 800 | 0.6779 | -1.3860 | -1.5069 | 0.5667 | 0.1209 | -377.0208 | -359.3108 | -4.2453 | -4.2572 |
| 0.6651 | 0.72 | 900 | 0.6819 | -1.6633 | -1.8040 | 0.5693 | 0.1408 | -406.7332 | -387.0414 | -4.6039 | -4.6178 |
| 0.5993 | 0.8 | 1000 | 0.6785 | -1.5776 | -1.7191 | 0.5683 | 0.1415 | -398.2364 | -378.4713 | -4.5356 | -4.5491 |
| 0.6759 | 0.88 | 1100 | 0.6778 | -1.5515 | -1.6923 | 0.5687 | 0.1408 | -395.5604 | -375.8654 | -4.4722 | -4.4855 |
| 0.6402 | 0.96 | 1200 | 0.6773 | -1.5439 | -1.6837 | 0.5690 | 0.1398 | -394.7002 | -375.1028 | -4.4444 | -4.4577 |
Framework versions
- PEFT 0.7.1
- Transformers 4.36.2
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.0