reflection_dpo.pt

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:

  • Logits/chosen: -0.9495
  • Logits/rejected: -0.9444
  • Logps/chosen: -75.1085
  • Logps/rejected: -153.4672
  • Loss: 0.2019
  • Rewards/accuracies: 0.9303
  • Rewards/chosen: -3.3185
  • Rewards/margins: 8.4758
  • Rewards/rejected: -11.7943

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: 0.0001
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • training_steps: 500

Training results

Training Loss Epoch Step Logits/chosen Logits/rejected Logps/chosen Logps/rejected Validation Loss Rewards/accuracies Rewards/chosen Rewards/margins Rewards/rejected
0.4496 0.0032 10 -0.7050 -0.6918 -44.4013 -46.5039 0.4352 0.8096 -0.2478 0.8502 -1.0980
0.3522 0.0064 20 -0.7595 -0.7500 -46.6875 -56.6043 0.3375 0.8405 -0.4764 1.6316 -2.1080
0.222 0.0096 30 -0.7949 -0.7888 -46.7048 -61.9410 0.2981 0.8520 -0.4782 2.1635 -2.6417
0.1372 0.0128 40 -0.8280 -0.8277 -43.7714 -64.4651 0.2727 0.8635 -0.1848 2.7093 -2.8941
0.3171 0.0160 50 -0.8491 -0.8518 -43.7808 -70.7219 0.2561 0.8685 -0.1858 3.3340 -3.5198
0.2395 0.0192 60 -0.8561 -0.8587 -44.3195 -74.5011 0.2392 0.8894 -0.2396 3.6581 -3.8977
0.1957 0.0224 70 -0.8841 -0.8821 -46.2857 -79.3155 0.2225 0.8958 -0.4363 3.9429 -4.3792
0.3734 0.0256 80 -0.9419 -0.9412 -48.2650 -82.6414 0.2188 0.8973 -0.6342 4.0776 -4.7118
0.1994 0.0288 90 -0.9986 -1.0003 -50.5246 -88.4025 0.2237 0.8980 -0.8602 4.4277 -5.2879
0.142 0.0320 100 -1.0438 -1.0429 -60.2203 -104.6738 0.2439 0.8901 -1.8297 5.0853 -6.9150
0.0637 0.0352 110 -1.0523 -1.0494 -64.6031 -112.8395 0.2623 0.8901 -2.2680 5.4636 -7.7316
0.3994 0.0384 120 -1.0401 -1.0326 -67.1203 -119.2853 0.2586 0.8930 -2.5197 5.8564 -8.3761
0.4963 0.0416 130 -1.0260 -1.0195 -63.6645 -114.9173 0.2443 0.8944 -2.1741 5.7652 -7.9393
0.184 0.0448 140 -0.9956 -0.9886 -64.3210 -116.1868 0.2440 0.8994 -2.2398 5.8265 -8.0663
0.4548 0.0480 150 -0.9475 -0.9363 -66.2748 -120.6504 0.2454 0.9016 -2.4352 6.0775 -8.5127
0.3672 0.0512 160 -0.9220 -0.9124 -62.2740 -113.5057 0.2140 0.9059 -2.0351 5.7631 -7.7982
0.1702 0.0544 170 -0.9642 -0.9637 -53.7962 -101.3066 0.1882 0.9167 -1.1873 5.3910 -6.5783
0.4943 0.0576 180 -0.9846 -0.9897 -49.7029 -94.7573 0.1848 0.9131 -0.7780 5.1454 -5.9233
0.4157 0.0608 190 -0.9938 -0.9997 -49.2515 -92.8510 0.1796 0.9159 -0.7329 4.9999 -5.7327
0.1773 0.0640 200 -1.0428 -1.0499 -50.9003 -96.4929 0.1829 0.9159 -0.8977 5.1992 -6.0969
0.061 0.0672 210 -1.0744 -1.0813 -53.2245 -101.3653 0.1894 0.9124 -1.1301 5.4540 -6.5841
0.2528 0.0704 220 -1.0751 -1.0807 -55.2461 -106.9378 0.1869 0.9174 -1.3323 5.8091 -7.1414
0.1233 0.0736 230 -1.0647 -1.0694 -58.8487 -115.0116 0.1922 0.9217 -1.6926 6.2562 -7.9488
0.102 0.0768 240 -1.0603 -1.0651 -60.8948 -118.7347 0.1928 0.9203 -1.8972 6.4239 -8.3211
0.3324 0.0800 250 -1.0533 -1.0557 -63.6693 -125.6155 0.1904 0.9224 -2.1746 6.8345 -9.0092
0.006 0.0832 260 -1.0321 -1.0304 -72.4092 -138.8671 0.2070 0.9195 -3.0486 7.2857 -10.3343
0.088 0.0864 270 -1.0164 -1.0114 -76.2556 -146.4741 0.2187 0.9203 -3.4333 7.6618 -11.0950
0.1346 0.0896 280 -0.9962 -0.9865 -80.2022 -149.6163 0.2305 0.9188 -3.8279 7.5813 -11.4092
0.4596 0.0928 290 -0.9952 -0.9845 -79.7203 -148.6709 0.2291 0.9210 -3.7797 7.5350 -11.3147
0.3019 0.0960 300 -1.0053 -0.9956 -78.3724 -147.5943 0.2222 0.9203 -3.6449 7.5621 -11.2071
0.0708 0.0992 310 -1.0132 -1.0058 -75.9911 -145.4509 0.2133 0.9188 -3.4068 7.5859 -10.9927
0.0371 0.1024 320 -1.0142 -1.0081 -79.0266 -152.0327 0.2291 0.9217 -3.7104 7.9405 -11.6509
0.3383 0.1056 330 -0.9949 -0.9877 -83.4509 -159.5245 0.2452 0.9167 -4.1528 8.2473 -12.4001
1.1015 0.1088 340 -0.9688 -0.9596 -84.9845 -163.4342 0.2508 0.9210 -4.3061 8.4849 -12.7910
0.2088 0.1120 350 -0.9577 -0.9474 -83.8031 -160.2732 0.2460 0.9181 -4.1880 8.2869 -12.4749
0.3555 0.1152 360 -0.9630 -0.9531 -80.5315 -156.9231 0.2314 0.9210 -3.8608 8.2791 -12.1399
0.197 0.1184 370 -0.9738 -0.9651 -78.5307 -154.6240 0.2215 0.9246 -3.6608 8.2493 -11.9100
0.5949 0.1216 380 -0.9840 -0.9770 -75.5271 -151.0730 0.2079 0.9260 -3.3604 8.1945 -11.5549
0.3272 0.1248 390 -0.9869 -0.9810 -74.7702 -150.4924 0.2008 0.9332 -3.2847 8.2121 -11.4969
0.0613 0.1280 400 -0.9777 -0.9725 -75.3109 -151.8670 0.1995 0.9325 -3.3388 8.2955 -11.6343
0.3468 0.1312 410 -0.9809 -0.9769 -73.3059 -148.0693 0.1949 0.9289 -3.1383 8.1163 -11.2545
0.4202 0.1344 420 -0.9788 -0.9744 -73.1894 -148.3277 0.1945 0.9274 -3.1266 8.1537 -11.2804
0.2138 0.1376 430 -0.9738 -0.9687 -73.5879 -149.2324 0.1947 0.9289 -3.1665 8.2044 -11.3709
0.475 0.1408 440 -0.9682 -0.9636 -73.3113 -149.3820 0.1959 0.9303 -3.1388 8.2470 -11.3858
0.0014 0.1440 450 -0.9672 -0.9625 -73.3271 -149.4055 0.1976 0.9296 -3.1404 8.2478 -11.3882
0.0216 0.1472 460 -0.9631 -0.9585 -73.6283 -150.2499 0.1994 0.9303 -3.1705 8.3021 -11.4726
0.1517 0.1504 470 -0.9563 -0.9517 -74.2343 -151.6451 0.2018 0.9303 -3.2311 8.3810 -11.6121
0.3719 0.1536 480 -0.9516 -0.9466 -74.7146 -152.7081 0.2016 0.9303 -3.2792 8.4393 -11.7184
0.1176 0.1567 490 -0.9502 -0.9450 -75.0118 -153.3125 0.2020 0.9296 -3.3089 8.4700 -11.7789
0.2333 0.1599 500 -0.9495 -0.9444 -75.1085 -153.4672 0.2019 0.9303 -3.3185 8.4758 -11.7943

Framework versions

  • PEFT 0.12.0
  • Transformers 4.44.2
  • Pytorch 2.4.0+cu124
  • Datasets 2.21.0
  • Tokenizers 0.19.1
Downloads last month
4
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for hlee131/Llama-3-OOD-Reflection-DPO

Adapter
(2905)
this model