metadata
library_name: transformers
license: llama3
base_model: tsavage68/IE_L3_1000steps_1e6rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: IE_L3_1000steps_1e6rate_05beta_cSFTDPO
results: []
IE_L3_1000steps_1e6rate_05beta_cSFTDPO
This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.1802
- Rewards/chosen: -1.4168
- Rewards/rejected: -13.8543
- Rewards/accuracies: 0.7400
- Rewards/margins: 12.4374
- Logps/rejected: -103.3358
- Logps/chosen: -85.6314
- Logits/rejected: -0.7970
- Logits/chosen: -0.7188
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.1906 | 0.4 | 50 | 0.1802 | -1.0109 | -11.1903 | 0.7400 | 10.1794 | -98.0078 | -84.8196 | -0.7939 | -0.7206 |
| 0.1386 | 0.8 | 100 | 0.1802 | -1.2190 | -12.1625 | 0.7400 | 10.9435 | -99.9523 | -85.2358 | -0.7944 | -0.7197 |
| 0.1386 | 1.2 | 150 | 0.1802 | -1.2782 | -12.5852 | 0.7400 | 11.3070 | -100.7976 | -85.3541 | -0.7943 | -0.7189 |
| 0.1733 | 1.6 | 200 | 0.1802 | -1.3094 | -13.0296 | 0.7400 | 11.7202 | -101.6864 | -85.4166 | -0.7948 | -0.7186 |
| 0.2253 | 2.0 | 250 | 0.1802 | -1.3248 | -13.1625 | 0.7400 | 11.8377 | -101.9522 | -85.4473 | -0.7952 | -0.7186 |
| 0.1386 | 2.4 | 300 | 0.1802 | -1.3337 | -13.2622 | 0.7400 | 11.9285 | -102.1515 | -85.4652 | -0.7942 | -0.7174 |
| 0.1213 | 2.8 | 350 | 0.1802 | -1.3670 | -13.4507 | 0.7400 | 12.0837 | -102.5286 | -85.5317 | -0.7953 | -0.7178 |
| 0.1906 | 3.2 | 400 | 0.1802 | -1.3818 | -13.5334 | 0.7400 | 12.1517 | -102.6941 | -85.5613 | -0.7964 | -0.7189 |
| 0.1906 | 3.6 | 450 | 0.1802 | -1.3800 | -13.5899 | 0.7400 | 12.2099 | -102.8071 | -85.5577 | -0.7964 | -0.7189 |
| 0.2079 | 4.0 | 500 | 0.1802 | -1.3816 | -13.6722 | 0.7400 | 12.2906 | -102.9716 | -85.5610 | -0.7966 | -0.7187 |
| 0.156 | 4.4 | 550 | 0.1802 | -1.4142 | -13.7800 | 0.7400 | 12.3657 | -103.1872 | -85.6262 | -0.7956 | -0.7175 |
| 0.1213 | 4.8 | 600 | 0.1802 | -1.3864 | -13.7736 | 0.7400 | 12.3872 | -103.1744 | -85.5705 | -0.7974 | -0.7192 |
| 0.1906 | 5.2 | 650 | 0.1802 | -1.4252 | -13.8450 | 0.7400 | 12.4197 | -103.3172 | -85.6483 | -0.7969 | -0.7187 |
| 0.2426 | 5.6 | 700 | 0.1802 | -1.4087 | -13.8154 | 0.7400 | 12.4068 | -103.2581 | -85.6151 | -0.7974 | -0.7196 |
| 0.2599 | 6.0 | 750 | 0.1802 | -1.4077 | -13.8712 | 0.7400 | 12.4635 | -103.3696 | -85.6131 | -0.7977 | -0.7194 |
| 0.1213 | 6.4 | 800 | 0.1802 | -1.4158 | -13.9034 | 0.7400 | 12.4876 | -103.4339 | -85.6293 | -0.7977 | -0.7195 |
| 0.2426 | 6.8 | 850 | 0.1802 | -1.4105 | -13.8922 | 0.7400 | 12.4817 | -103.4116 | -85.6187 | -0.7979 | -0.7200 |
| 0.1733 | 7.2 | 900 | 0.1802 | -1.4075 | -13.8657 | 0.7400 | 12.4582 | -103.3587 | -85.6128 | -0.7970 | -0.7189 |
| 0.1386 | 7.6 | 950 | 0.1802 | -1.4138 | -13.8523 | 0.7400 | 12.4386 | -103.3319 | -85.6253 | -0.7971 | -0.7188 |
| 0.156 | 8.0 | 1000 | 0.1802 | -1.4168 | -13.8543 | 0.7400 | 12.4374 | -103.3358 | -85.6314 | -0.7970 | -0.7188 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.0.0+cu117
- Datasets 3.0.0
- Tokenizers 0.19.1