zephyr-7b-dpo-full
This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0254
- Rewards/chosen: -6.9910
- Rewards/rejected: -27.5001
- Rewards/accuracies: 0.9806
- Rewards/margins: 20.5091
- Logps/rejected: -345.1052
- Logps/chosen: -161.1468
- Logits/rejected: -1.1372
- Logits/chosen: -1.3943
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-07
- train_batch_size: 8
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 64
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.3601 | 0.05 | 100 | 0.3410 | -0.0714 | -1.0669 | 0.9444 | 0.9955 | -80.7736 | -91.9507 | -3.0201 | -3.0686 |
| 0.113 | 0.09 | 200 | 0.1171 | -1.2736 | -4.8848 | 0.9611 | 3.6112 | -118.9524 | -103.9727 | -3.0104 | -3.0297 |
| 0.0734 | 0.14 | 300 | 0.0768 | -2.2058 | -7.2581 | 0.9778 | 5.0523 | -142.6854 | -113.2948 | -2.9044 | -2.9330 |
| 0.0587 | 0.18 | 400 | 0.0559 | -4.0388 | -11.7401 | 0.9694 | 7.7012 | -187.5052 | -131.6255 | -2.6966 | -2.7279 |
| 0.0379 | 0.23 | 500 | 0.0460 | -3.9501 | -12.2735 | 0.9750 | 8.3233 | -192.8393 | -130.7384 | -2.6613 | -2.7017 |
| 0.0394 | 0.27 | 600 | 0.0405 | -4.8526 | -14.4947 | 0.9750 | 9.6420 | -215.0514 | -139.7636 | -2.3901 | -2.4786 |
| 0.0375 | 0.32 | 700 | 0.0376 | -3.7095 | -12.6204 | 0.9750 | 8.9110 | -196.3089 | -128.3319 | -2.6708 | -2.7142 |
| 0.043 | 0.37 | 800 | 0.0375 | -5.2652 | -14.8540 | 0.9694 | 9.5888 | -218.6447 | -143.8889 | -2.1654 | -2.2829 |
| 0.0304 | 0.41 | 900 | 0.0380 | -4.2196 | -13.7402 | 0.9750 | 9.5207 | -207.5071 | -133.4332 | -2.1041 | -2.2015 |
| 0.0254 | 0.46 | 1000 | 0.0324 | -4.9741 | -16.1886 | 0.9722 | 11.2145 | -231.9906 | -140.9781 | -2.0860 | -2.1849 |
| 0.03 | 0.5 | 1100 | 0.0342 | -5.1530 | -16.4077 | 0.9667 | 11.2547 | -234.1812 | -142.7671 | -2.1132 | -2.1749 |
| 0.0339 | 0.55 | 1200 | 0.0311 | -2.7641 | -11.9997 | 0.9750 | 9.2356 | -190.1020 | -118.8785 | -2.0003 | -2.1125 |
| 0.0489 | 0.59 | 1300 | 0.0272 | -2.8374 | -12.4331 | 0.9833 | 9.5957 | -194.4359 | -119.6116 | -2.1739 | -2.2572 |
| 0.0263 | 0.64 | 1400 | 0.0291 | -3.4940 | -13.2533 | 0.9833 | 9.7593 | -202.6373 | -126.1769 | -2.0814 | -2.2103 |
| 0.0301 | 0.68 | 1500 | 0.0266 | -4.7000 | -16.2144 | 0.9778 | 11.5144 | -232.2482 | -138.2372 | -2.0580 | -2.1682 |
| 0.0272 | 0.73 | 1600 | 0.0283 | -6.1020 | -18.1286 | 0.9667 | 12.0266 | -251.3911 | -152.2577 | -1.8091 | -1.9654 |
| 0.0278 | 0.78 | 1700 | 0.0254 | -2.9913 | -13.0756 | 0.9750 | 10.0843 | -200.8606 | -121.1501 | -2.1657 | -2.2443 |
| 0.0291 | 0.82 | 1800 | 0.0260 | -4.8989 | -16.4318 | 0.9722 | 11.5329 | -234.4226 | -140.2258 | -1.9427 | -2.0364 |
| 0.0253 | 0.87 | 1900 | 0.0252 | -4.0403 | -15.3217 | 0.9778 | 11.2813 | -223.3212 | -131.6404 | -1.9099 | -2.0361 |
| 0.0235 | 0.91 | 2000 | 0.0223 | -3.1973 | -14.8592 | 0.9750 | 11.6619 | -218.6964 | -123.2101 | -2.0448 | -2.1573 |
| 0.0272 | 0.96 | 2100 | 0.0236 | -3.5828 | -15.3281 | 0.9750 | 11.7453 | -223.3855 | -127.0649 | -1.9834 | -2.0892 |
| 0.0107 | 1.0 | 2200 | 0.0206 | -2.6028 | -14.3743 | 0.9806 | 11.7715 | -213.8473 | -117.2654 | -2.1607 | -2.2658 |
| 0.0089 | 1.05 | 2300 | 0.0209 | -4.2378 | -18.4889 | 0.9806 | 14.2511 | -254.9935 | -133.6152 | -1.8951 | -2.0551 |
| 0.0103 | 1.1 | 2400 | 0.0222 | -3.8267 | -16.7762 | 0.9778 | 12.9495 | -237.8671 | -129.5042 | -2.1516 | -2.2577 |
| 0.0114 | 1.14 | 2500 | 0.0230 | -5.6573 | -20.9027 | 0.9750 | 15.2454 | -279.1315 | -147.8097 | -1.6603 | -1.8552 |
| 0.0075 | 1.19 | 2600 | 0.0217 | -4.9252 | -19.3893 | 0.9833 | 14.4641 | -263.9972 | -140.4892 | -1.8500 | -2.0172 |
| 0.0084 | 1.23 | 2700 | 0.0235 | -5.3971 | -20.1630 | 0.9806 | 14.7660 | -271.7348 | -145.2077 | -1.8673 | -2.0178 |
| 0.0094 | 1.28 | 2800 | 0.0253 | -4.4422 | -18.2275 | 0.9778 | 13.7853 | -252.3796 | -135.6592 | -2.0417 | -2.1522 |
| 0.0064 | 1.32 | 2900 | 0.0269 | -4.3177 | -18.7842 | 0.9750 | 14.4665 | -257.9463 | -134.4144 | -2.0273 | -2.1416 |
| 0.0093 | 1.37 | 3000 | 0.0234 | -4.6335 | -19.3959 | 0.9722 | 14.7624 | -264.0636 | -137.5719 | -1.9133 | -2.0781 |
| 0.0083 | 1.41 | 3100 | 0.0230 | -5.0798 | -20.1230 | 0.9806 | 15.0431 | -271.3341 | -142.0356 | -1.7624 | -1.9098 |
| 0.0108 | 1.46 | 3200 | 0.0217 | -3.8920 | -18.1320 | 0.9806 | 14.2400 | -251.4249 | -130.1573 | -1.9500 | -2.0745 |
| 0.0091 | 1.51 | 3300 | 0.0223 | -5.6969 | -21.8594 | 0.9806 | 16.1624 | -288.6983 | -148.2067 | -1.9011 | -2.0357 |
| 0.012 | 1.55 | 3400 | 0.0215 | -4.8539 | -18.2010 | 0.9861 | 13.3471 | -252.1147 | -139.7764 | -1.7467 | -1.8729 |
| 0.0064 | 1.6 | 3500 | 0.0218 | -5.0782 | -19.6405 | 0.9833 | 14.5623 | -266.5100 | -142.0197 | -1.7721 | -1.8976 |
| 0.0078 | 1.64 | 3600 | 0.0213 | -5.1302 | -20.2433 | 0.9778 | 15.1130 | -272.5372 | -142.5395 | -1.5605 | -1.7238 |
| 0.0076 | 1.69 | 3700 | 0.0222 | -5.6579 | -22.5669 | 0.9778 | 16.9089 | -295.7735 | -147.8167 | -1.4705 | -1.6559 |
| 0.0124 | 1.73 | 3800 | 0.0221 | -3.5477 | -16.1322 | 0.9861 | 12.5845 | -231.4264 | -126.7141 | -1.6382 | -1.7606 |
| 0.0065 | 1.78 | 3900 | 0.0201 | -5.0615 | -20.0072 | 0.9806 | 14.9457 | -270.1767 | -141.8525 | -1.5657 | -1.7132 |
| 0.0078 | 1.83 | 4000 | 0.0200 | -5.5288 | -20.5404 | 0.9806 | 15.0115 | -275.5083 | -146.5255 | -1.6180 | -1.7627 |
| 0.0089 | 1.87 | 4100 | 0.0210 | -5.5847 | -22.2109 | 0.9806 | 16.6262 | -292.2132 | -147.0840 | -1.6373 | -1.7998 |
| 0.0022 | 1.92 | 4200 | 0.0206 | -5.2113 | -21.4282 | 0.9833 | 16.2169 | -284.3861 | -143.3500 | -1.5687 | -1.7379 |
| 0.0065 | 1.96 | 4300 | 0.0205 | -4.4336 | -18.8360 | 0.9806 | 14.4024 | -258.4650 | -135.5732 | -1.5889 | -1.7431 |
| 0.0038 | 2.01 | 4400 | 0.0213 | -3.8031 | -17.3943 | 0.9833 | 13.5912 | -244.0473 | -129.2682 | -1.6138 | -1.7573 |
| 0.0042 | 2.05 | 4500 | 0.0210 | -4.7797 | -20.6519 | 0.9833 | 15.8722 | -276.6233 | -139.0337 | -1.4571 | -1.6393 |
| 0.0025 | 2.1 | 4600 | 0.0220 | -5.3030 | -22.1802 | 0.9833 | 16.8772 | -291.9064 | -144.2672 | -1.4136 | -1.6114 |
| 0.0013 | 2.15 | 4700 | 0.0240 | -6.9006 | -26.5257 | 0.9806 | 19.6251 | -335.3617 | -160.2434 | -1.3102 | -1.5345 |
| 0.0075 | 2.19 | 4800 | 0.0253 | -6.6613 | -26.8086 | 0.9778 | 20.1473 | -338.1903 | -157.8497 | -1.2594 | -1.4933 |
| 0.0024 | 2.24 | 4900 | 0.0238 | -5.9384 | -25.2963 | 0.9806 | 19.3579 | -323.0674 | -150.6208 | -1.3039 | -1.5318 |
| 0.0017 | 2.28 | 5000 | 0.0217 | -5.7146 | -23.2521 | 0.9806 | 17.5375 | -302.6257 | -148.3833 | -1.3104 | -1.5277 |
| 0.0037 | 2.33 | 5100 | 0.0234 | -6.8711 | -26.8591 | 0.9806 | 19.9880 | -338.6961 | -159.9486 | -1.2368 | -1.4798 |
| 0.0045 | 2.37 | 5200 | 0.0233 | -6.4563 | -25.4117 | 0.9806 | 18.9554 | -324.2215 | -155.7999 | -1.2791 | -1.5095 |
| 0.0024 | 2.42 | 5300 | 0.0226 | -6.0528 | -23.6960 | 0.9806 | 17.6432 | -307.0642 | -151.7647 | -1.2664 | -1.4908 |
| 0.0023 | 2.46 | 5400 | 0.0233 | -6.6659 | -25.6664 | 0.9806 | 19.0004 | -326.7684 | -157.8966 | -1.2361 | -1.4788 |
| 0.0007 | 2.51 | 5500 | 0.0250 | -6.7756 | -26.1454 | 0.9806 | 19.3698 | -331.5591 | -158.9936 | -1.2610 | -1.4948 |
| 0.0029 | 2.56 | 5600 | 0.0246 | -6.6468 | -25.9933 | 0.9806 | 19.3465 | -330.0380 | -157.7054 | -1.3105 | -1.5473 |
| 0.0017 | 2.6 | 5700 | 0.0248 | -6.9095 | -26.0328 | 0.9806 | 19.1234 | -330.4331 | -160.3320 | -1.2445 | -1.4880 |
| 0.0055 | 2.65 | 5800 | 0.0257 | -7.5666 | -27.8213 | 0.9806 | 20.2547 | -348.3180 | -166.9033 | -1.1526 | -1.4106 |
| 0.0036 | 2.69 | 5900 | 0.0263 | -7.6427 | -28.5222 | 0.9806 | 20.8795 | -355.3266 | -167.6644 | -1.0877 | -1.3563 |
| 0.0061 | 2.74 | 6000 | 0.0242 | -6.7938 | -26.3053 | 0.9806 | 19.5115 | -333.1572 | -159.1749 | -1.2176 | -1.4633 |
| 0.0018 | 2.78 | 6100 | 0.0242 | -6.8675 | -26.5397 | 0.9806 | 19.6722 | -335.5020 | -159.9126 | -1.1958 | -1.4429 |
| 0.0024 | 2.83 | 6200 | 0.0249 | -7.1136 | -27.2224 | 0.9806 | 20.1088 | -342.3287 | -162.3729 | -1.1466 | -1.4003 |
| 0.0026 | 2.88 | 6300 | 0.0251 | -6.9595 | -27.0923 | 0.9806 | 20.1328 | -341.0272 | -160.8322 | -1.1501 | -1.4025 |
| 0.0034 | 2.92 | 6400 | 0.0253 | -7.1263 | -27.7050 | 0.9806 | 20.5787 | -347.1544 | -162.5005 | -1.1288 | -1.3867 |
| 0.0023 | 2.97 | 6500 | 0.0253 | -7.0828 | -27.6979 | 0.9806 | 20.6151 | -347.0838 | -162.0652 | -1.1284 | -1.3867 |
Framework versions
- Transformers 4.35.0
- Pytorch 2.1.1+cu121
- Datasets 2.14.6
- Tokenizers 0.14.1
- Downloads last month
- 3
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Model tree for nlile/zephyr-7b-dpo-full
Base model
mistralai/Mistral-7B-v0.1
Finetuned
alignment-handbook/zephyr-7b-sft-full