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_1e7rate_03beta_cSFTDPO
results: []
IE_L3_1000steps_1e7rate_03beta_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: -1.0922
- Rewards/rejected: -10.0336
- Rewards/accuracies: 0.7400
- Rewards/margins: 8.9414
- Logps/rejected: -109.0726
- Logps/chosen: -86.4386
- Logits/rejected: -0.8003
- Logits/chosen: -0.7150
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-07
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5177 | 0.4 | 50 | 0.4502 | -0.0545 | -0.6708 | 0.7400 | 0.6163 | -77.8632 | -82.9793 | -0.7959 | -0.7377 |
| 0.1418 | 0.8 | 100 | 0.1816 | -0.5946 | -5.5441 | 0.7400 | 4.9495 | -94.1076 | -84.7799 | -0.7931 | -0.7224 |
| 0.1388 | 1.2 | 150 | 0.1803 | -0.8790 | -8.0921 | 0.7400 | 7.2131 | -102.6009 | -85.7278 | -0.7962 | -0.7173 |
| 0.1733 | 1.6 | 200 | 0.1803 | -0.9325 | -8.7008 | 0.7400 | 7.7683 | -104.6298 | -85.9060 | -0.7980 | -0.7170 |
| 0.2253 | 2.0 | 250 | 0.1803 | -0.9653 | -8.8898 | 0.7400 | 7.9244 | -105.2598 | -86.0156 | -0.7979 | -0.7163 |
| 0.1387 | 2.4 | 300 | 0.1802 | -0.9837 | -9.1362 | 0.7400 | 8.1525 | -106.0812 | -86.0766 | -0.7975 | -0.7157 |
| 0.1213 | 2.8 | 350 | 0.1802 | -1.0210 | -9.4276 | 0.7400 | 8.4066 | -107.0527 | -86.2011 | -0.7989 | -0.7159 |
| 0.1906 | 3.2 | 400 | 0.1802 | -1.0245 | -9.5511 | 0.7400 | 8.5265 | -107.4642 | -86.2129 | -0.7991 | -0.7152 |
| 0.1906 | 3.6 | 450 | 0.1802 | -1.0419 | -9.6482 | 0.7400 | 8.6063 | -107.7879 | -86.2706 | -0.7995 | -0.7155 |
| 0.208 | 4.0 | 500 | 0.1802 | -1.0676 | -9.8319 | 0.7400 | 8.7643 | -108.4001 | -86.3564 | -0.7999 | -0.7153 |
| 0.156 | 4.4 | 550 | 0.1802 | -1.0697 | -9.9071 | 0.7400 | 8.8374 | -108.6509 | -86.3635 | -0.8011 | -0.7160 |
| 0.1213 | 4.8 | 600 | 0.1802 | -1.0716 | -9.9151 | 0.7400 | 8.8436 | -108.6776 | -86.3697 | -0.8002 | -0.7154 |
| 0.1906 | 5.2 | 650 | 0.1802 | -1.0758 | -9.9883 | 0.7400 | 8.9125 | -108.9217 | -86.3839 | -0.8005 | -0.7154 |
| 0.2426 | 5.6 | 700 | 0.1802 | -1.0847 | -10.0383 | 0.7400 | 8.9536 | -109.0882 | -86.4134 | -0.8003 | -0.7150 |
| 0.2599 | 6.0 | 750 | 0.1802 | -1.0957 | -10.0559 | 0.7400 | 8.9602 | -109.1469 | -86.4500 | -0.8008 | -0.7156 |
| 0.1213 | 6.4 | 800 | 0.1802 | -1.0865 | -10.0490 | 0.7400 | 8.9625 | -109.1239 | -86.4195 | -0.7997 | -0.7139 |
| 0.2426 | 6.8 | 850 | 0.1802 | -1.0859 | -10.0366 | 0.7400 | 8.9506 | -109.0825 | -86.4176 | -0.8000 | -0.7146 |
| 0.1733 | 7.2 | 900 | 0.1802 | -1.0860 | -10.0398 | 0.7400 | 8.9538 | -109.0932 | -86.4178 | -0.8002 | -0.7149 |
| 0.1386 | 7.6 | 950 | 0.1802 | -1.0922 | -10.0336 | 0.7400 | 8.9414 | -109.0726 | -86.4386 | -0.8003 | -0.7150 |
| 0.156 | 8.0 | 1000 | 0.1802 | -1.0922 | -10.0336 | 0.7400 | 8.9414 | -109.0726 | -86.4386 | -0.8003 | -0.7150 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.0.0+cu117
- Datasets 3.0.0
- Tokenizers 0.19.1