metadata
license: apache-2.0
base_model: mosaicml/mpt-7b-instruct
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: MPT_1000_STEPS_1e7_rate_03_beta_DPO
results: []
MPT_1000_STEPS_1e7_rate_03_beta_DPO
This model is a fine-tuned version of mosaicml/mpt-7b-instruct on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6919
- Rewards/chosen: -0.0230
- Rewards/rejected: -0.0291
- Rewards/accuracies: 0.5275
- Rewards/margins: 0.0061
- Logps/rejected: -21.6156
- Logps/chosen: -20.8382
- Logits/rejected: 14.2213
- Logits/chosen: 14.2239
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6958 | 0.05 | 50 | 0.6969 | -0.0103 | -0.0064 | 0.4791 | -0.0040 | -21.5702 | -20.8128 | 14.2683 | 14.2709 |
| 0.6948 | 0.1 | 100 | 0.6966 | -0.0023 | 0.0014 | 0.5077 | -0.0037 | -21.5546 | -20.7968 | 14.2571 | 14.2597 |
| 0.6971 | 0.15 | 150 | 0.7007 | -0.0051 | 0.0067 | 0.4681 | -0.0117 | -21.5441 | -20.8024 | 14.2475 | 14.2501 |
| 0.6891 | 0.2 | 200 | 0.6943 | 0.0187 | 0.0174 | 0.4923 | 0.0013 | -21.5227 | -20.7548 | 14.2452 | 14.2478 |
| 0.6906 | 0.24 | 250 | 0.6922 | 0.0036 | -0.0018 | 0.4747 | 0.0054 | -21.5609 | -20.7850 | 14.2395 | 14.2421 |
| 0.6865 | 0.29 | 300 | 0.6942 | 0.0038 | 0.0023 | 0.4857 | 0.0015 | -21.5528 | -20.7845 | 14.2393 | 14.2419 |
| 0.7058 | 0.34 | 350 | 0.6939 | -0.0025 | -0.0045 | 0.5055 | 0.0020 | -21.5664 | -20.7971 | 14.2533 | 14.2559 |
| 0.6817 | 0.39 | 400 | 0.6918 | -0.0255 | -0.0318 | 0.5143 | 0.0063 | -21.6210 | -20.8431 | 14.2343 | 14.2369 |
| 0.6726 | 0.44 | 450 | 0.6902 | -0.0203 | -0.0301 | 0.5582 | 0.0099 | -21.6177 | -20.8327 | 14.2287 | 14.2313 |
| 0.6927 | 0.49 | 500 | 0.6903 | -0.0159 | -0.0254 | 0.5209 | 0.0096 | -21.6083 | -20.8239 | 14.2329 | 14.2355 |
| 0.6728 | 0.54 | 550 | 0.6905 | -0.0252 | -0.0342 | 0.5297 | 0.0089 | -21.6258 | -20.8426 | 14.2305 | 14.2331 |
| 0.6733 | 0.59 | 600 | 0.6877 | -0.0158 | -0.0305 | 0.5341 | 0.0147 | -21.6184 | -20.8237 | 14.2330 | 14.2356 |
| 0.6937 | 0.64 | 650 | 0.6916 | -0.0222 | -0.0293 | 0.5341 | 0.0071 | -21.6161 | -20.8365 | 14.2242 | 14.2268 |
| 0.6771 | 0.68 | 700 | 0.6921 | -0.0234 | -0.0294 | 0.5231 | 0.0060 | -21.6163 | -20.8391 | 14.2289 | 14.2315 |
| 0.6874 | 0.73 | 750 | 0.6916 | -0.0219 | -0.0286 | 0.5121 | 0.0067 | -21.6147 | -20.8361 | 14.2292 | 14.2317 |
| 0.6772 | 0.78 | 800 | 0.6888 | -0.0187 | -0.0313 | 0.5473 | 0.0127 | -21.6201 | -20.8295 | 14.2308 | 14.2334 |
| 0.7033 | 0.83 | 850 | 0.6886 | -0.0163 | -0.0294 | 0.5297 | 0.0131 | -21.6163 | -20.8248 | 14.2220 | 14.2245 |
| 0.6772 | 0.88 | 900 | 0.6894 | -0.0217 | -0.0330 | 0.5297 | 0.0113 | -21.6235 | -20.8357 | 14.2227 | 14.2253 |
| 0.696 | 0.93 | 950 | 0.6918 | -0.0229 | -0.0293 | 0.5275 | 0.0064 | -21.6160 | -20.8380 | 14.2213 | 14.2239 |
| 0.6881 | 0.98 | 1000 | 0.6919 | -0.0230 | -0.0291 | 0.5275 | 0.0061 | -21.6156 | -20.8382 | 14.2213 | 14.2239 |
Framework versions
- Transformers 4.39.1
- Pytorch 2.0.0+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2