llama-7b-SFT-qlora-wiki_DPO_ds_RM_contrast_1024_r_64_alpha_16
This model is a fine-tuned version of dhmeltzer/llama-7b-SFT_ds_wiki65k_1024_r_64_alpha_16_merged on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6249
- Rewards/chosen: -0.1145
- Rewards/rejected: -0.3663
- Rewards/accuracies: 0.6451
- Rewards/margins: 0.2518
- Logps/rejected: -202.5918
- Logps/chosen: -208.1973
- Logits/rejected: 1.1131
- Logits/chosen: 1.1385
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: 0.0002
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.03
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.692 | 0.1 | 19 | 0.6510 | -0.3841 | -0.5657 | 0.5949 | 0.1816 | -204.5857 | -210.8932 | 1.1059 | 1.1283 |
| 0.6585 | 0.21 | 38 | 0.6389 | -0.0095 | -0.2372 | 0.6373 | 0.2276 | -201.3002 | -207.1476 | 1.1111 | 1.1363 |
| 0.6581 | 0.31 | 57 | 0.6299 | -0.0360 | -0.3003 | 0.6417 | 0.2643 | -201.9318 | -207.4127 | 1.1053 | 1.1315 |
| 0.6485 | 0.42 | 76 | 0.6332 | -0.2261 | -0.4511 | 0.6194 | 0.2250 | -203.4390 | -209.3129 | 1.0905 | 1.1138 |
| 0.6551 | 0.52 | 95 | 0.6270 | -0.1240 | -0.3577 | 0.6362 | 0.2337 | -202.5053 | -208.2919 | 1.1088 | 1.1331 |
| 0.6484 | 0.62 | 114 | 0.6293 | -0.1372 | -0.3680 | 0.6440 | 0.2308 | -202.6089 | -208.4242 | 1.1213 | 1.1467 |
| 0.6427 | 0.73 | 133 | 0.6264 | -0.1804 | -0.4360 | 0.6451 | 0.2556 | -203.2879 | -208.8561 | 1.1096 | 1.1347 |
| 0.645 | 0.83 | 152 | 0.6249 | -0.1145 | -0.3663 | 0.6451 | 0.2518 | -202.5918 | -208.1973 | 1.1131 | 1.1385 |
| 0.6335 | 0.94 | 171 | 0.6253 | -0.1000 | -0.3475 | 0.6339 | 0.2476 | -202.4036 | -208.0517 | 1.1149 | 1.1394 |
Framework versions
- Transformers 4.32.1
- Pytorch 2.0.1+cu118
- Datasets 2.14.4
- Tokenizers 0.13.3