ultra-feedback-dutch-cleaned-hq-spin-geitje-7b-ultra-sft_iter0

This model is a fine-tuned version of BramVanroy/GEITje-7B-ultra-sft on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0135
  • Rewards/real: -1.4818
  • Rewards/generated: -13.3376
  • Rewards/accuracies: 0.9963
  • Rewards/margins: 11.8558
  • Logps/generated: -410.0757
  • Logps/real: -427.4978
  • Logits/generated: -2.7305
  • Logits/real: -2.7643

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: 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
0.4944 0.08 25 0.2566 0.6645 -0.8427 0.9761 1.5071 -285.1264 -406.0350 -3.0069 -3.0147
0.092 0.16 50 0.0838 0.3983 -3.7771 0.9890 4.1754 -314.4705 -408.6964 -2.9427 -2.9557
0.0601 0.25 75 0.0457 0.2564 -5.6388 0.9963 5.8952 -333.0871 -410.1154 -2.9205 -2.9326
0.0437 0.33 100 0.0336 -0.1853 -7.2451 0.9963 7.0598 -349.1503 -414.5328 -2.8883 -2.9062
0.036 0.41 125 0.0271 -0.1651 -7.7408 0.9945 7.5756 -354.1071 -414.3309 -2.8817 -2.9014
0.0373 0.49 150 0.0264 -0.2384 -7.8312 0.9908 7.5928 -355.0117 -415.0634 -2.8271 -2.8543
0.0198 0.58 175 0.0214 -0.9152 -9.9469 0.9908 9.0317 -376.1681 -421.8315 -2.8052 -2.8326
0.0426 0.66 200 0.0251 -0.9747 -9.1022 0.9908 8.1275 -367.7210 -422.4266 -2.8450 -2.8588
0.0262 0.74 225 0.0189 -0.8414 -9.9318 0.9926 9.0903 -376.0172 -421.0940 -2.8009 -2.8209
0.0142 0.82 250 0.0166 -0.7154 -10.1059 0.9945 9.3905 -377.7586 -419.8336 -2.7973 -2.8201
0.0171 0.9 275 0.0189 -1.0905 -10.9057 0.9945 9.8151 -385.7561 -423.5849 -2.7641 -2.7936
0.0333 0.99 300 0.0168 -1.2797 -11.4866 0.9963 10.2069 -391.5655 -425.4765 -2.7973 -2.8230
0.0061 1.07 325 0.0157 -1.2079 -11.1880 0.9945 9.9801 -388.5797 -424.7587 -2.7974 -2.8231
0.0022 1.15 350 0.0152 -1.0695 -11.2438 0.9908 10.1743 -389.1376 -423.3746 -2.7853 -2.8128
0.0033 1.23 375 0.0148 -1.1767 -11.6618 0.9908 10.4851 -393.3175 -424.4465 -2.7751 -2.8029
0.0043 1.32 400 0.0138 -1.0951 -11.8306 0.9963 10.7354 -395.0049 -423.6307 -2.7703 -2.7976
0.005 1.4 425 0.0136 -1.3179 -12.4674 0.9963 11.1494 -401.3733 -425.8589 -2.7573 -2.7851
0.0031 1.48 450 0.0139 -1.3771 -12.6901 0.9963 11.3130 -403.6003 -426.4503 -2.7544 -2.7815
0.0039 1.56 475 0.0134 -1.3885 -12.8092 0.9963 11.4207 -404.7912 -426.5648 -2.7446 -2.7735
0.001 1.64 500 0.0136 -1.4378 -13.0038 0.9963 11.5660 -406.7370 -427.0571 -2.7404 -2.7701
0.0059 1.73 525 0.0139 -1.5924 -13.4168 0.9945 11.8244 -410.8671 -428.6035 -2.7293 -2.7629
0.0015 1.81 550 0.0136 -1.5136 -13.3984 0.9963 11.8848 -410.6832 -427.8157 -2.7283 -2.7623
0.0078 1.89 575 0.0135 -1.4891 -13.3323 0.9963 11.8432 -410.0224 -427.5704 -2.7309 -2.7645
0.0043 1.97 600 0.0135 -1.4818 -13.3376 0.9963 11.8558 -410.0757 -427.4978 -2.7305 -2.7643

Framework versions

  • Transformers 4.37.0
  • Pytorch 2.1.2+cu121
  • Datasets 2.14.6
  • Tokenizers 0.15.2
Downloads last month
32
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_iter0

Finetuned
(3)
this model
Finetunes
1 model