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_01_beta_DPO
results: []
v1_1000_STEPS_1e8_rate_01_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: 0.6930
- Rewards/chosen: -0.0000
- Rewards/rejected: -0.0004
- Rewards/accuracies: 0.4571
- Rewards/margins: 0.0004
- Logps/rejected: -16.8832
- Logps/chosen: -15.2532
- Logits/rejected: -3.3537
- Logits/chosen: -3.3538
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6937 | 0.05 | 50 | 0.6931 | -0.0004 | -0.0006 | 0.4813 | 0.0002 | -16.8854 | -15.2569 | -3.3537 | -3.3538 |
| 0.6934 | 0.1 | 100 | 0.6930 | 0.0001 | -0.0003 | 0.4879 | 0.0004 | -16.8825 | -15.2519 | -3.3538 | -3.3539 |
| 0.6945 | 0.15 | 150 | 0.6931 | -0.0002 | -0.0003 | 0.4725 | 0.0001 | -16.8826 | -15.2550 | -3.3538 | -3.3539 |
| 0.6945 | 0.2 | 200 | 0.6933 | -0.0003 | -0.0000 | 0.4396 | -0.0003 | -16.8800 | -15.2561 | -3.3537 | -3.3538 |
| 0.693 | 0.24 | 250 | 0.6933 | -0.0002 | 0.0000 | 0.4352 | -0.0002 | -16.8791 | -15.2548 | -3.3538 | -3.3539 |
| 0.6934 | 0.29 | 300 | 0.6930 | 0.0003 | -0.0001 | 0.4549 | 0.0004 | -16.8806 | -15.2500 | -3.3539 | -3.3540 |
| 0.6945 | 0.34 | 350 | 0.6931 | -0.0007 | -0.0007 | 0.4462 | 0.0001 | -16.8867 | -15.2596 | -3.3539 | -3.3539 |
| 0.6914 | 0.39 | 400 | 0.6933 | -0.0006 | -0.0004 | 0.4593 | -0.0002 | -16.8834 | -15.2590 | -3.3538 | -3.3538 |
| 0.693 | 0.44 | 450 | 0.6932 | -0.0001 | -0.0001 | 0.4396 | -0.0000 | -16.8802 | -15.2539 | -3.3537 | -3.3537 |
| 0.6937 | 0.49 | 500 | 0.6931 | -0.0002 | -0.0003 | 0.4484 | 0.0001 | -16.8827 | -15.2548 | -3.3538 | -3.3539 |
| 0.6918 | 0.54 | 550 | 0.6930 | -0.0001 | -0.0004 | 0.4791 | 0.0003 | -16.8834 | -15.2541 | -3.3537 | -3.3538 |
| 0.6926 | 0.59 | 600 | 0.6933 | -0.0005 | -0.0002 | 0.4505 | -0.0002 | -16.8820 | -15.2580 | -3.3537 | -3.3537 |
| 0.6922 | 0.64 | 650 | 0.6933 | -0.0003 | -0.0000 | 0.4549 | -0.0003 | -16.8796 | -15.2563 | -3.3538 | -3.3539 |
| 0.693 | 0.68 | 700 | 0.6931 | -0.0006 | -0.0007 | 0.4484 | 0.0001 | -16.8861 | -15.2589 | -3.3538 | -3.3539 |
| 0.6926 | 0.73 | 750 | 0.6928 | 0.0001 | -0.0006 | 0.4879 | 0.0007 | -16.8858 | -15.2520 | -3.3538 | -3.3539 |
| 0.693 | 0.78 | 800 | 0.6931 | -0.0004 | -0.0007 | 0.4923 | 0.0002 | -16.8861 | -15.2574 | -3.3538 | -3.3539 |
| 0.693 | 0.83 | 850 | 0.6930 | -0.0002 | -0.0005 | 0.4462 | 0.0003 | -16.8842 | -15.2548 | -3.3537 | -3.3538 |
| 0.6926 | 0.88 | 900 | 0.6930 | -0.0000 | -0.0004 | 0.4571 | 0.0004 | -16.8832 | -15.2532 | -3.3537 | -3.3538 |
| 0.6937 | 0.93 | 950 | 0.6930 | -0.0000 | -0.0004 | 0.4571 | 0.0004 | -16.8832 | -15.2532 | -3.3537 | -3.3538 |
| 0.6926 | 0.98 | 1000 | 0.6930 | -0.0000 | -0.0004 | 0.4571 | 0.0004 | -16.8832 | -15.2532 | -3.3537 | -3.3538 |
Framework versions
- Transformers 4.39.1
- Pytorch 2.0.0+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2