metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-Instruct-v0.1
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: v1_1000_STEPS_1e8_rate_03_beta_DPO
results: []
v1_1000_STEPS_1e8_rate_03_beta_DPO
This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6933
- Rewards/chosen: -0.0011
- Rewards/rejected: -0.0010
- Rewards/accuracies: 0.4615
- Rewards/margins: -0.0001
- Logps/rejected: -16.8827
- Logps/chosen: -15.2567
- Logits/rejected: -3.3538
- Logits/chosen: -3.3539
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: 1e-08
- train_batch_size: 2
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 1000
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6951 | 0.05 | 50 | 0.6937 | -0.0020 | -0.0011 | 0.4681 | -0.0009 | -16.8832 | -15.2599 | -3.3538 | -3.3539 |
| 0.6947 | 0.1 | 100 | 0.6933 | 0.0007 | 0.0008 | 0.4484 | -0.0001 | -16.8769 | -15.2508 | -3.3538 | -3.3538 |
| 0.6922 | 0.15 | 150 | 0.6932 | 0.0005 | 0.0004 | 0.4615 | 0.0001 | -16.8783 | -15.2513 | -3.3537 | -3.3538 |
| 0.6927 | 0.2 | 200 | 0.6937 | -0.0002 | 0.0007 | 0.4527 | -0.0009 | -16.8771 | -15.2536 | -3.3538 | -3.3538 |
| 0.6924 | 0.24 | 250 | 0.6927 | 0.0006 | -0.0005 | 0.4593 | 0.0011 | -16.8811 | -15.2510 | -3.3537 | -3.3538 |
| 0.6916 | 0.29 | 300 | 0.6934 | -0.0007 | -0.0004 | 0.4418 | -0.0003 | -16.8810 | -15.2554 | -3.3538 | -3.3538 |
| 0.6948 | 0.34 | 350 | 0.6932 | 0.0004 | 0.0003 | 0.4637 | 0.0001 | -16.8785 | -15.2516 | -3.3538 | -3.3539 |
| 0.6929 | 0.39 | 400 | 0.6925 | -0.0004 | -0.0018 | 0.4637 | 0.0014 | -16.8855 | -15.2543 | -3.3538 | -3.3538 |
| 0.69 | 0.44 | 450 | 0.6936 | -0.0007 | -0.0000 | 0.4374 | -0.0007 | -16.8796 | -15.2555 | -3.3536 | -3.3537 |
| 0.694 | 0.49 | 500 | 0.6930 | -0.0004 | -0.0008 | 0.4505 | 0.0005 | -16.8823 | -15.2542 | -3.3538 | -3.3538 |
| 0.6895 | 0.54 | 550 | 0.6932 | -0.0009 | -0.0012 | 0.4703 | 0.0002 | -16.8834 | -15.2562 | -3.3537 | -3.3537 |
| 0.6955 | 0.59 | 600 | 0.6930 | 0.0007 | 0.0002 | 0.4747 | 0.0004 | -16.8788 | -15.2509 | -3.3538 | -3.3539 |
| 0.6903 | 0.64 | 650 | 0.6934 | -0.0005 | -0.0003 | 0.4593 | -0.0002 | -16.8804 | -15.2548 | -3.3537 | -3.3538 |
| 0.6904 | 0.68 | 700 | 0.6934 | -0.0004 | -0.0001 | 0.4549 | -0.0003 | -16.8800 | -15.2544 | -3.3538 | -3.3538 |
| 0.6921 | 0.73 | 750 | 0.6930 | -0.0008 | -0.0013 | 0.4703 | 0.0004 | -16.8838 | -15.2558 | -3.3538 | -3.3539 |
| 0.6945 | 0.78 | 800 | 0.6930 | -0.0003 | -0.0008 | 0.4813 | 0.0005 | -16.8823 | -15.2540 | -3.3538 | -3.3539 |
| 0.6915 | 0.83 | 850 | 0.6939 | -0.0016 | -0.0003 | 0.4484 | -0.0014 | -16.8804 | -15.2585 | -3.3538 | -3.3539 |
| 0.6903 | 0.88 | 900 | 0.6933 | -0.0011 | -0.0010 | 0.4615 | -0.0001 | -16.8827 | -15.2567 | -3.3538 | -3.3539 |
| 0.6971 | 0.93 | 950 | 0.6933 | -0.0011 | -0.0010 | 0.4615 | -0.0001 | -16.8827 | -15.2567 | -3.3538 | -3.3539 |
| 0.6939 | 0.98 | 1000 | 0.6933 | -0.0011 | -0.0010 | 0.4615 | -0.0001 | -16.8827 | -15.2567 | -3.3538 | -3.3539 |
Framework versions
- Transformers 4.39.1
- Pytorch 2.0.0+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2