metadata
license: llama3
base_model: tsavage68/MedQA_L3_1000steps_1e6rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: MedQA_L3_1000steps_1e6rate_01beta_CSFTDPO
results: []
MedQA_L3_1000steps_1e6rate_01beta_CSFTDPO
This model is a fine-tuned version of tsavage68/MedQA_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.4143
- Rewards/chosen: -0.2461
- Rewards/rejected: -2.6298
- Rewards/accuracies: 0.8088
- Rewards/margins: 2.3838
- Logps/rejected: -60.1531
- Logps/chosen: -33.7891
- Logits/rejected: -1.3940
- Logits/chosen: -1.3910
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-06
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6869 | 0.0489 | 50 | 0.6696 | -0.2211 | -0.2710 | 0.7253 | 0.0498 | -36.5645 | -33.5400 | -0.7298 | -0.7290 |
| 0.4779 | 0.0977 | 100 | 0.5887 | 1.4526 | 1.0417 | 0.6945 | 0.4109 | -23.4374 | -16.8024 | -0.8047 | -0.8036 |
| 0.5752 | 0.1466 | 150 | 0.4975 | 0.5331 | -0.2997 | 0.7473 | 0.8328 | -36.8518 | -25.9976 | -0.8723 | -0.8705 |
| 0.4157 | 0.1954 | 200 | 0.5087 | -0.0815 | -1.0065 | 0.7538 | 0.9250 | -43.9199 | -32.1434 | -0.9039 | -0.9019 |
| 0.4271 | 0.2443 | 250 | 0.4619 | 0.5202 | -0.5333 | 0.7648 | 1.0535 | -39.1874 | -26.1265 | -0.9341 | -0.9319 |
| 0.3162 | 0.2931 | 300 | 0.4272 | 0.2052 | -1.3157 | 0.8110 | 1.5209 | -47.0122 | -29.2765 | -1.0303 | -1.0281 |
| 0.3868 | 0.3420 | 350 | 0.4366 | 0.0191 | -1.4354 | 0.7868 | 1.4545 | -48.2090 | -31.1376 | -1.1172 | -1.1146 |
| 0.4267 | 0.3908 | 400 | 0.4253 | 0.8142 | -0.6501 | 0.8044 | 1.4642 | -40.3556 | -23.1869 | -1.2091 | -1.2069 |
| 0.4816 | 0.4397 | 450 | 0.4235 | 0.7057 | -0.6954 | 0.7978 | 1.4011 | -40.8093 | -24.2719 | -1.2618 | -1.2590 |
| 0.5777 | 0.4885 | 500 | 0.4147 | 0.5199 | -1.2061 | 0.8088 | 1.7260 | -45.9158 | -26.1293 | -1.3148 | -1.3119 |
| 0.3051 | 0.5374 | 550 | 0.4133 | 0.2933 | -1.3715 | 0.8022 | 1.6647 | -47.5694 | -28.3956 | -1.3646 | -1.3616 |
| 0.5378 | 0.5862 | 600 | 0.4219 | -0.4403 | -2.6925 | 0.8088 | 2.2522 | -60.7803 | -35.7319 | -1.3525 | -1.3496 |
| 0.359 | 0.6351 | 650 | 0.4122 | -0.0585 | -2.2242 | 0.8132 | 2.1656 | -56.0965 | -31.9139 | -1.3793 | -1.3763 |
| 0.4137 | 0.6839 | 700 | 0.4019 | 0.0561 | -2.0220 | 0.8066 | 2.0781 | -54.0746 | -30.7675 | -1.3921 | -1.3890 |
| 0.3899 | 0.7328 | 750 | 0.4093 | -0.1488 | -2.4231 | 0.8110 | 2.2743 | -58.0863 | -32.8165 | -1.3920 | -1.3890 |
| 0.3645 | 0.7816 | 800 | 0.4095 | -0.2104 | -2.5505 | 0.8132 | 2.3401 | -59.3594 | -33.4322 | -1.3965 | -1.3935 |
| 0.4993 | 0.8305 | 850 | 0.4157 | -0.2412 | -2.6172 | 0.8088 | 2.3760 | -60.0272 | -33.7410 | -1.3947 | -1.3918 |
| 0.6907 | 0.8793 | 900 | 0.4164 | -0.2462 | -2.6292 | 0.8110 | 2.3829 | -60.1466 | -33.7908 | -1.3944 | -1.3914 |
| 0.3846 | 0.9282 | 950 | 0.4140 | -0.2447 | -2.6315 | 0.8110 | 2.3868 | -60.1702 | -33.7755 | -1.3939 | -1.3909 |
| 0.3404 | 0.9770 | 1000 | 0.4143 | -0.2461 | -2.6298 | 0.8088 | 2.3838 | -60.1531 | -33.7891 | -1.3940 | -1.3910 |
Framework versions
- Transformers 4.41.1
- Pytorch 2.0.0+cu117
- Datasets 2.19.1
- Tokenizers 0.19.1