metadata
base_model: mistralai/Mistral-7B-v0.1
library_name: peft
license: apache-2.0
tags:
- alignment-handbook
- trl
- dpo
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-lora
results: []
zephyr-7b-dpo-lora
This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-qlora on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4930
- Rewards/chosen: -1.7956
- Rewards/rejected: -2.7390
- Rewards/accuracies: 0.7242
- Rewards/margins: 0.9434
- Logps/rejected: -536.2667
- Logps/chosen: -447.0530
- Logits/rejected: 0.9396
- Logits/chosen: 0.5316
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
- num_devices: 4
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- total_eval_batch_size: 32
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6087 | 0.1 | 100 | 0.6158 | -0.3136 | -0.5466 | 0.6726 | 0.2330 | -317.0252 | -298.8513 | -2.0360 | -2.1198 |
| 0.5463 | 0.21 | 200 | 0.5504 | -1.1262 | -1.6978 | 0.6925 | 0.5716 | -432.1413 | -380.1157 | -0.0431 | -0.2986 |
| 0.4949 | 0.31 | 300 | 0.5161 | -1.6535 | -2.4330 | 0.7183 | 0.7794 | -505.6621 | -432.8479 | 0.4034 | 0.1418 |
| 0.5239 | 0.42 | 400 | 0.5101 | -1.3693 | -2.0810 | 0.7302 | 0.7116 | -470.4624 | -404.4282 | 0.8585 | 0.5591 |
| 0.5272 | 0.52 | 500 | 0.5003 | -2.0358 | -2.9629 | 0.7381 | 0.9271 | -558.6534 | -471.0703 | 1.0404 | 0.7150 |
| 0.4886 | 0.63 | 600 | 0.4982 | -1.7739 | -2.6428 | 0.7262 | 0.8689 | -526.6414 | -444.8822 | 0.3752 | 0.0594 |
| 0.516 | 0.73 | 700 | 0.4933 | -2.0243 | -2.9388 | 0.7302 | 0.9144 | -556.2413 | -469.9273 | 0.8898 | 0.5312 |
| 0.495 | 0.84 | 800 | 0.4949 | -1.7382 | -2.6840 | 0.7262 | 0.9458 | -530.7620 | -441.3121 | 0.8308 | 0.4157 |
| 0.4866 | 0.94 | 900 | 0.4932 | -1.7916 | -2.7322 | 0.7262 | 0.9407 | -535.5854 | -446.6503 | 0.9353 | 0.5257 |
Framework versions
- PEFT 0.7.1
- Transformers 4.38.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.15.2