Llama-2-7b-hf-DPO-LookAhead3_FullEval_TTree1.4_TLoop0.7_TEval0.2_Filter0.2_V1.0
This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.9877
- Rewards/chosen: -2.3284
- Rewards/rejected: -2.4506
- Rewards/accuracies: 0.5
- Rewards/margins: 0.1222
- Logps/rejected: -132.7494
- Logps/chosen: -125.7301
- Logits/rejected: -0.6876
- Logits/chosen: -0.6598
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: 5e-05
- train_batch_size: 2
- eval_batch_size: 2
- 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: 10
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6798 | 0.3021 | 50 | 0.7124 | -0.0006 | 0.0284 | 0.5 | -0.0290 | -107.9596 | -102.4519 | -0.1526 | -0.1158 |
| 0.622 | 0.6042 | 100 | 0.6927 | 0.0459 | 0.0462 | 0.375 | -0.0004 | -107.7811 | -101.9873 | -0.1568 | -0.1192 |
| 0.6031 | 0.9063 | 150 | 0.6815 | 0.0644 | 0.0476 | 0.375 | 0.0168 | -107.7677 | -101.8019 | -0.2034 | -0.1651 |
| 0.3606 | 1.2085 | 200 | 0.8146 | -0.7471 | -0.7226 | 0.625 | -0.0245 | -115.4695 | -109.9166 | -0.3703 | -0.3356 |
| 0.3387 | 1.5106 | 250 | 0.6641 | -0.3875 | -0.5323 | 0.625 | 0.1448 | -113.5663 | -106.3212 | -0.3313 | -0.2957 |
| 0.1549 | 1.8127 | 300 | 0.6263 | -0.8093 | -1.0444 | 0.625 | 0.2351 | -118.6870 | -110.5388 | -0.3892 | -0.3537 |
| 0.0958 | 2.1148 | 350 | 0.7394 | -1.7451 | -2.0072 | 0.5 | 0.2621 | -128.3158 | -119.8970 | -0.5348 | -0.5043 |
| 0.0193 | 2.4169 | 400 | 0.9249 | -2.0984 | -2.2555 | 0.5 | 0.1571 | -130.7979 | -123.4299 | -0.6495 | -0.6219 |
| 0.3616 | 2.7190 | 450 | 0.9877 | -2.3284 | -2.4506 | 0.5 | 0.1222 | -132.7494 | -125.7301 | -0.6876 | -0.6598 |
Framework versions
- PEFT 0.12.0
- Transformers 4.44.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.0
- Tokenizers 0.19.1
- Downloads last month
- 1
Model tree for LBK95/Llama-2-7b-hf-DPO-LookAhead3_FullEval_TTree1.4_TLoop0.7_TEval0.2_Filter0.2_V1.0
Base model
meta-llama/Llama-2-7b-hf