phi-2-dpo-full-hydrox-safe

This model is a fine-tuned version of microsoft/phi-2 on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0050
  • Rewards/chosen: -0.0996
  • Rewards/rejected: -19.3989
  • Rewards/accuracies: 0.9992
  • Rewards/margins: 19.2993
  • Logps/rejected: -644.0859
  • Logps/chosen: -275.4780
  • Logits/rejected: 0.2021
  • Logits/chosen: -0.3196

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
0.6535 0.03 100 0.6404 0.0912 -0.0433 0.6936 0.1345 -450.5297 -273.5692 0.6266 0.1056
0.4293 0.07 200 0.4021 0.4957 -0.4032 0.8998 0.8989 -454.1287 -269.5248 0.6425 0.1325
0.2685 0.1 300 0.2464 0.8729 -1.0624 0.9386 1.9354 -460.7211 -265.7526 0.6700 0.1533
0.1918 0.14 400 0.1792 1.1981 -1.7756 0.9453 2.9737 -467.8528 -262.5003 0.6918 0.1719
0.1409 0.17 500 0.1423 1.4079 -2.4222 0.9503 3.8301 -474.3184 -260.4028 0.7139 0.1772
0.1437 0.2 600 0.1232 1.6054 -3.1009 0.9537 4.7063 -481.1060 -258.4279 0.7179 0.1571
0.0747 0.24 700 0.1057 1.7190 -3.9721 0.9613 5.6911 -489.8179 -257.2914 0.7179 0.1325
0.0848 0.27 800 0.0863 1.8540 -4.3289 0.9689 6.1829 -493.3858 -255.9415 0.7047 0.1024
0.0853 0.31 900 0.0744 1.8328 -5.0270 0.9722 6.8598 -500.3666 -256.1535 0.6746 0.0529
0.0561 0.34 1000 0.0639 1.7736 -5.4555 0.9747 7.2291 -504.6516 -256.7455 0.6517 0.0185
0.0657 0.37 1100 0.0546 1.7966 -5.6453 0.9840 7.4419 -506.5500 -256.5157 0.6738 0.0168
0.0445 0.41 1200 0.0499 1.7245 -6.0645 0.9823 7.7889 -510.7414 -257.2369 0.6757 0.0176
0.0498 0.44 1300 0.0459 1.6804 -6.6799 0.9848 8.3603 -516.8956 -257.6772 0.6719 0.0228
0.0554 0.48 1400 0.0549 1.5994 -6.8731 0.9848 8.4726 -518.8279 -258.4872 0.6684 0.0151
0.0368 0.51 1500 0.0365 1.3551 -7.8016 0.9865 9.1567 -528.1130 -260.9311 0.6464 -0.0223
0.0514 0.54 1600 0.0333 1.3431 -7.5785 0.9891 8.9215 -525.8817 -261.0511 0.6617 -0.0159
0.0365 0.58 1700 0.0310 1.0743 -8.8130 0.9933 9.8873 -538.2264 -263.7386 0.6354 -0.0234
0.0214 0.61 1800 0.0300 1.1310 -9.1633 0.9924 10.2942 -541.7292 -263.1721 0.6278 -0.0199
0.0321 0.65 1900 0.0267 0.9514 -9.9884 0.9924 10.9399 -549.9811 -264.9675 0.6044 -0.0268
0.0239 0.68 2000 0.0249 0.9624 -10.0608 0.9933 11.0232 -550.7047 -264.8574 0.5955 -0.0628
0.0484 0.71 2100 0.0222 0.9176 -11.0294 0.9941 11.9470 -560.3909 -265.3055 0.6043 -0.0338
0.0239 0.75 2200 0.0209 0.8247 -10.5070 0.9941 11.3317 -555.1663 -266.2345 0.5913 -0.0648
0.0219 0.78 2300 0.0221 0.7334 -10.5131 0.9924 11.2465 -555.2272 -267.1474 0.5942 -0.0808
0.0168 0.82 2400 0.0200 0.8909 -11.0818 0.9966 11.9727 -560.9146 -265.5724 0.6006 -0.0770
0.0135 0.85 2500 0.0187 0.7273 -11.3130 0.9958 12.0403 -563.2269 -267.2084 0.5948 -0.0711
0.0211 0.88 2600 0.0190 0.5390 -11.8492 0.9949 12.3883 -568.5890 -269.0913 0.5818 -0.0719
0.0209 0.92 2700 0.0170 0.4957 -12.3176 0.9949 12.8133 -573.2725 -269.5249 0.5537 -0.0912
0.0168 0.95 2800 0.0175 0.5850 -12.2737 0.9958 12.8587 -572.8340 -268.6321 0.5471 -0.0870
0.0129 0.99 2900 0.0171 0.5691 -12.7785 0.9949 13.3476 -577.8817 -268.7910 0.5441 -0.0810
0.0134 1.02 3000 0.0188 0.3713 -13.0254 0.9966 13.3966 -580.3504 -270.7691 0.5261 -0.1163
0.0127 1.05 3100 0.0155 0.1769 -14.3013 0.9958 14.4782 -593.1094 -272.7126 0.4928 -0.1129
0.0072 1.09 3200 0.0147 0.2474 -13.6499 0.9975 13.8973 -586.5952 -272.0074 0.4753 -0.1479
0.0068 1.12 3300 0.0179 0.2168 -14.6122 0.9958 14.8291 -596.2192 -272.3131 0.4722 -0.1430
0.0323 1.16 3400 0.0218 0.2536 -15.9479 0.9958 16.2015 -609.5762 -271.9459 0.4159 -0.1430
0.0232 1.19 3500 0.0171 0.4273 -14.2304 0.9958 14.6577 -592.4003 -270.2084 0.4598 -0.1410
0.0112 1.22 3600 0.0165 0.3634 -15.1069 0.9966 15.4704 -601.1661 -270.8472 0.4366 -0.1521
0.0136 1.26 3700 0.0176 0.2922 -14.4784 0.9966 14.7706 -594.8809 -271.5597 0.4553 -0.1679
0.0089 1.29 3800 0.0155 0.2349 -14.8970 0.9966 15.1319 -599.0669 -272.1331 0.4447 -0.1721
0.01 1.33 3900 0.0123 0.1309 -15.3620 0.9975 15.4930 -603.7170 -273.1724 0.4180 -0.1834
0.0055 1.36 4000 0.0132 0.0227 -16.5037 0.9975 16.5265 -615.1339 -274.2542 0.3836 -0.1899
0.004 1.39 4100 0.0122 0.2637 -15.2111 0.9975 15.4748 -602.2075 -271.8445 0.4087 -0.1870
0.0057 1.43 4200 0.0117 0.1593 -16.5411 0.9975 16.7003 -615.5074 -272.8891 0.4086 -0.1518
0.0041 1.46 4300 0.0113 -0.0815 -17.4813 0.9975 17.3998 -624.9092 -275.2963 0.3722 -0.1782
0.0039 1.5 4400 0.0115 0.2051 -17.4309 0.9983 17.6359 -624.4056 -272.4310 0.3445 -0.1975
0.0075 1.53 4500 0.0098 0.2810 -15.5168 0.9992 15.7978 -605.2650 -271.6719 0.4021 -0.1980
0.0034 1.56 4600 0.0123 0.2926 -15.4764 0.9992 15.7690 -604.8610 -271.5560 0.4225 -0.1766
0.0066 1.6 4700 0.0125 0.3192 -14.8325 0.9992 15.1517 -598.4222 -271.2900 0.4068 -0.2026
0.0079 1.63 4800 0.0124 0.1405 -15.1594 0.9992 15.2999 -601.6904 -273.0765 0.3942 -0.2089
0.0063 1.67 4900 0.0108 -0.1407 -17.4715 0.9975 17.3308 -624.8116 -275.8886 0.3294 -0.2140
0.0026 1.7 5000 0.0116 -0.0520 -17.9554 0.9992 17.9034 -629.6504 -275.0013 0.3361 -0.2062
0.0151 1.73 5100 0.0112 0.5727 -14.3495 0.9975 14.9221 -593.5914 -268.7550 0.4234 -0.1897
0.0049 1.77 5200 0.0083 0.2709 -15.7593 0.9975 16.0301 -607.6893 -271.7730 0.3788 -0.2054
0.0036 1.8 5300 0.0093 0.2336 -16.3789 0.9992 16.6125 -613.8856 -272.1456 0.3798 -0.1961
0.0076 1.84 5400 0.0086 0.2618 -16.1654 0.9992 16.4272 -611.7511 -271.8639 0.3844 -0.2090
0.0028 1.87 5500 0.0079 0.3444 -15.7817 0.9983 16.1261 -607.9135 -271.0378 0.3874 -0.2135
0.0102 1.9 5600 0.0081 0.3276 -15.7265 0.9983 16.0541 -607.3622 -271.2061 0.3713 -0.2243
0.0051 1.94 5700 0.0072 0.3979 -15.8670 0.9983 16.2649 -608.7670 -270.5030 0.3562 -0.2429
0.0068 1.97 5800 0.0069 0.2817 -16.7894 0.9992 17.0711 -617.9911 -271.6647 0.3329 -0.2424
0.0021 2.01 5900 0.0061 0.0725 -17.3567 0.9992 17.4292 -623.6634 -273.7564 0.3142 -0.2479
0.0031 2.04 6000 0.0066 -0.0179 -18.1905 0.9983 18.1726 -632.0013 -274.6606 0.2948 -0.2445
0.0032 2.07 6100 0.0064 0.0140 -17.8441 0.9992 17.8581 -628.5374 -274.3414 0.3043 -0.2424
0.0043 2.11 6200 0.0066 0.0216 -18.5824 0.9983 18.6041 -635.9210 -274.2653 0.2863 -0.2416
0.0033 2.14 6300 0.0070 0.0355 -18.3245 0.9983 18.3600 -633.3419 -274.1270 0.2783 -0.2691
0.0036 2.18 6400 0.0065 0.1473 -18.2292 0.9983 18.3765 -632.3882 -273.0083 0.2772 -0.2744
0.0023 2.21 6500 0.0064 0.0450 -17.9052 0.9992 17.9502 -629.1483 -274.0311 0.2750 -0.2756
0.0025 2.24 6600 0.0061 -0.0642 -17.4265 0.9983 17.3623 -624.3619 -275.1237 0.2773 -0.2881
0.0008 2.28 6700 0.0058 0.0951 -17.3710 0.9992 17.4661 -623.8071 -273.5309 0.2703 -0.2891
0.002 2.31 6800 0.0057 -0.0533 -18.2744 0.9992 18.2212 -632.8410 -275.0142 0.2544 -0.2823
0.0052 2.35 6900 0.0073 0.0152 -18.6276 0.9983 18.6427 -636.3723 -274.3301 0.2365 -0.2928
0.0029 2.38 7000 0.0056 0.0321 -18.4600 0.9992 18.4921 -634.6968 -274.1606 0.2366 -0.2889
0.0013 2.41 7100 0.0058 -0.1304 -19.1740 0.9983 19.0436 -641.8367 -275.7859 0.2306 -0.2898
0.0011 2.45 7200 0.0059 -0.0977 -19.2699 0.9992 19.1722 -642.7958 -275.4590 0.2227 -0.2916
0.0058 2.48 7300 0.0059 -0.0938 -18.8625 0.9992 18.7687 -638.7216 -275.4195 0.2314 -0.2945
0.004 2.52 7400 0.0067 -0.1457 -19.2264 0.9992 19.0807 -642.3610 -275.9389 0.2194 -0.2959
0.0042 2.55 7500 0.0056 -0.0100 -18.6009 0.9992 18.5909 -636.1058 -274.5812 0.2398 -0.3003
0.0033 2.58 7600 0.0057 -0.1484 -19.3114 0.9992 19.1630 -643.2104 -275.9655 0.2215 -0.2965
0.0012 2.62 7700 0.0051 -0.1574 -19.2398 0.9992 19.0824 -642.4946 -276.0552 0.2210 -0.3052
0.002 2.65 7800 0.0052 -0.1220 -19.5223 0.9992 19.4004 -645.3201 -275.7011 0.2102 -0.3165
0.0053 2.69 7900 0.0051 -0.1063 -19.6889 0.9992 19.5826 -646.9856 -275.5444 0.2106 -0.3074
0.0013 2.72 8000 0.0056 -0.1357 -19.8339 0.9992 19.6982 -648.4356 -275.8386 0.2029 -0.3129
0.002 2.75 8100 0.0050 -0.1024 -19.6476 0.9992 19.5452 -646.5726 -275.5056 0.1993 -0.3168
0.0016 2.79 8200 0.0052 -0.0378 -19.4058 0.9992 19.3680 -644.1544 -274.8596 0.2152 -0.3120
0.0015 2.82 8300 0.0049 -0.0549 -19.2544 0.9992 19.1994 -642.6404 -275.0310 0.2093 -0.3168
0.003 2.86 8400 0.0051 -0.1218 -19.2614 0.9983 19.1396 -642.7106 -275.7000 0.2057 -0.3182
0.0019 2.89 8500 0.0050 -0.1057 -19.3041 0.9992 19.1983 -643.1373 -275.5387 0.2077 -0.3151
0.0039 2.92 8600 0.0052 -0.0863 -19.2826 0.9992 19.1963 -642.9229 -275.3445 0.2057 -0.3208
0.0042 2.96 8700 0.0048 -0.1111 -19.3837 0.9992 19.2726 -643.9339 -275.5926 0.2008 -0.3215
0.0023 2.99 8800 0.0050 -0.1085 -19.4076 0.9992 19.2990 -644.1723 -275.5668 0.1985 -0.3228

Framework versions

  • Transformers 4.38.0.dev0
  • Pytorch 2.1.1+cu121
  • Datasets 2.14.6
  • Tokenizers 0.15.1
Downloads last month
22
Safetensors
Model size
2.78B params
Tensor type
BF16
·
Inference Examples
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 yihang7/phi-2-dpo-full-hydrox-safe

Base model

microsoft/phi-2
Finetuned
(287)
this model