ultra-feedback-dutch-cleaned-hq-spin-geitje-7b-ultra-sft_iter2

This model is a fine-tuned version of davidberenstein1957/ultra-feedback-dutch-cleaned-hq-spin-geitje-7b-ultra-sft_iter1 on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0162
  • Rewards/real: -8.1731
  • Rewards/generated: -31.3826
  • Rewards/accuracies: 0.9917
  • Rewards/margins: 23.2095
  • Logps/generated: -956.3063
  • Logps/real: -525.1735
  • Logits/generated: -1.5719
  • Logits/real: -1.7813

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: 8
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • gradient_accumulation_steps: 2
  • 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: 2

Training results

Training Loss Epoch Step Validation Loss Rewards/real Rewards/generated Rewards/accuracies Rewards/margins Logps/generated Logps/real Logits/generated Logits/real
0.6097 0.04 25 0.4147 -0.6192 -1.4312 0.9250 0.8120 -656.7919 -449.6341 -2.0004 -2.0773
0.2137 0.08 50 0.1745 -2.0300 -5.0060 0.9519 2.9761 -692.5404 -463.7422 -1.9306 -2.0237
0.1292 0.12 75 0.1012 -2.8227 -7.4967 0.9685 4.6740 -717.4471 -471.6697 -1.8843 -1.9887
0.0665 0.16 100 0.0676 -3.2936 -9.3177 0.9778 6.0240 -735.6567 -476.3786 -1.8508 -1.9628
0.0429 0.21 125 0.0477 -3.7328 -11.2722 0.9824 7.5395 -755.2025 -480.7701 -1.8123 -1.9332
0.0299 0.25 150 0.0369 -4.2161 -13.2599 0.9870 9.0437 -775.0787 -485.6039 -1.7938 -1.9226
0.0252 0.29 175 0.0320 -4.7201 -15.0489 0.9880 10.3288 -792.9691 -490.6432 -1.7758 -1.9116
0.0249 0.33 200 0.0301 -5.0757 -16.3570 0.9880 11.2813 -806.0497 -494.1995 -1.7515 -1.8923
0.0175 0.37 225 0.0273 -5.4299 -17.6751 0.9880 12.2451 -819.2310 -497.7419 -1.7362 -1.8821
0.0183 0.41 250 0.0254 -5.4183 -18.3899 0.9889 12.9715 -826.3791 -497.6259 -1.7300 -1.8793
0.0182 0.45 275 0.0245 -6.0900 -20.5760 0.9889 14.4860 -848.2401 -504.3426 -1.6961 -1.8564
0.0253 0.49 300 0.0224 -5.9239 -20.7184 0.9898 14.7944 -849.6640 -502.6819 -1.6938 -1.8573
0.0075 0.53 325 0.0234 -7.0436 -24.1126 0.9898 17.0691 -883.6064 -513.8781 -1.6522 -1.8252
0.0141 0.58 350 0.0212 -5.5696 -20.9714 0.9898 15.4017 -852.1937 -499.1387 -1.7082 -1.8693
0.0135 0.62 375 0.0182 -5.2646 -20.3901 0.9907 15.1254 -846.3809 -496.0890 -1.7285 -1.8897
0.014 0.66 400 0.0182 -5.5057 -21.1579 0.9907 15.6522 -854.0594 -498.4994 -1.7137 -1.8783
0.0122 0.7 425 0.0172 -5.3398 -20.7520 0.9907 15.4122 -849.9997 -496.8405 -1.7231 -1.8857
0.0144 0.74 450 0.0164 -4.6606 -19.3766 0.9917 14.7160 -836.2463 -490.0483 -1.7465 -1.9042
0.0103 0.78 475 0.0160 -4.8739 -20.1058 0.9907 15.2319 -843.5385 -492.1819 -1.7445 -1.9064
0.0147 0.82 500 0.0156 -5.1220 -20.9607 0.9917 15.8387 -852.0875 -494.6623 -1.7434 -1.9092
0.0154 0.86 525 0.0155 -5.1481 -21.3994 0.9917 16.2513 -856.4740 -494.9235 -1.7357 -1.9040
0.0158 0.91 550 0.0151 -5.6088 -22.9532 0.9917 17.3444 -872.0123 -499.5304 -1.7139 -1.8881
0.0053 0.95 575 0.0149 -5.7209 -23.5217 0.9917 17.8008 -877.6972 -500.6515 -1.7113 -1.8888
0.008 0.99 600 0.0147 -5.7523 -23.7474 0.9917 17.9952 -879.9544 -500.9651 -1.7086 -1.8878
0.0049 1.03 625 0.0154 -6.1839 -24.8883 0.9907 18.7044 -891.3632 -505.2818 -1.6731 -1.8585
0.0057 1.07 650 0.0155 -6.4947 -25.8924 0.9917 19.3977 -901.4037 -508.3892 -1.6592 -1.8484
0.0076 1.11 675 0.0158 -6.8543 -26.9217 0.9917 20.0674 -911.6970 -511.9859 -1.6407 -1.8339
0.004 1.15 700 0.0158 -7.1325 -27.7743 0.9917 20.6418 -920.2236 -514.7678 -1.6269 -1.8236
0.0168 1.19 725 0.0157 -6.9019 -26.2791 0.9917 19.3772 -905.2711 -512.4611 -1.6566 -1.8448
0.0022 1.23 750 0.0163 -6.9586 -26.5145 0.9917 19.5559 -907.6251 -513.0281 -1.6533 -1.8423
0.0039 1.28 775 0.0165 -7.5386 -28.2224 0.9917 20.6837 -924.7038 -518.8289 -1.6369 -1.8327
0.002 1.32 800 0.0165 -7.6568 -28.6441 0.9907 20.9872 -928.9208 -520.0109 -1.6365 -1.8344
0.002 1.36 825 0.0165 -7.7989 -29.2028 0.9917 21.4038 -934.5078 -521.4318 -1.6348 -1.8352
0.0019 1.4 850 0.0165 -7.8978 -29.5958 0.9917 21.6980 -938.4382 -522.4203 -1.6166 -1.8169
0.0041 1.44 875 0.0162 -7.9696 -29.7930 0.9917 21.8234 -940.4100 -523.1380 -1.6165 -1.8176
0.0023 1.48 900 0.0164 -8.2086 -30.6909 0.9917 22.4823 -949.3892 -525.5286 -1.6045 -1.8093
0.0038 1.52 925 0.0166 -8.1217 -30.6727 0.9917 22.5510 -949.2076 -524.6597 -1.5919 -1.7978
0.0096 1.56 950 0.0162 -7.8257 -30.1144 0.9917 22.2887 -943.6237 -521.6992 -1.5909 -1.7956
0.0057 1.6 975 0.0166 -8.0335 -30.6654 0.9917 22.6319 -949.1342 -523.7775 -1.5854 -1.7919
0.0046 1.65 1000 0.0165 -8.1757 -31.0139 0.9917 22.8382 -952.6191 -525.2000 -1.5768 -1.7852
0.0009 1.69 1025 0.0165 -8.0553 -30.7565 0.9917 22.7012 -950.0453 -523.9951 -1.5757 -1.7830
0.002 1.73 1050 0.0164 -8.1838 -31.3365 0.9917 23.1528 -955.8453 -525.2800 -1.5692 -1.7790
0.0069 1.77 1075 0.0163 -8.1908 -31.4118 0.9917 23.2210 -956.5981 -525.3508 -1.5749 -1.7850
0.0029 1.81 1100 0.0166 -8.4138 -32.0830 0.9917 23.6692 -963.3098 -527.5802 -1.5624 -1.7752
0.0047 1.85 1125 0.0166 -8.4223 -32.1526 0.9917 23.7304 -964.0065 -527.6652 -1.5631 -1.7759
0.0037 1.89 1150 0.0163 -8.1563 -31.3209 0.9917 23.1646 -955.6895 -525.0057 -1.5739 -1.7832
0.0026 1.93 1175 0.0163 -8.2107 -31.5009 0.9917 23.2901 -957.4888 -525.5498 -1.5708 -1.7807
0.0058 1.98 1200 0.0162 -8.1731 -31.3826 0.9917 23.2095 -956.3063 -525.1735 -1.5719 -1.7813

Framework versions

  • Transformers 4.37.0
  • Pytorch 2.1.2+cu121
  • Datasets 2.14.6
  • Tokenizers 0.15.2
Downloads last month
18
Safetensors
Model size
7.24B 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 davidberenstein1957/ultra-feedback-dutch-cleaned-hq-spin-geitje-7b-ultra-sft_iter2