model_usp4_dpo1

This model is a fine-tuned version of meta-llama/Llama-2-7b-chat-hf on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 1.3096
  • Rewards/chosen: -11.2358
  • Rewards/rejected: -13.1040
  • Rewards/accuracies: 0.5700
  • Rewards/margins: 1.8682
  • Logps/rejected: -241.2410
  • Logps/chosen: -223.0633
  • Logits/rejected: -1.1809
  • Logits/chosen: -1.2295

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: 0.0005
  • train_batch_size: 4
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 100
  • training_steps: 1000

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.098 2.67 100 0.9024 -5.2759 -6.1963 0.6200 0.9203 -172.1640 -163.4645 -1.3943 -1.4049
0.014 5.33 200 1.0817 -5.6325 -6.8881 0.5900 1.2556 -179.0825 -167.0302 -1.3979 -1.4111
0.0002 8.0 300 1.2922 -11.0538 -12.8527 0.5700 1.7989 -238.7282 -221.2436 -1.1960 -1.2430
0.0001 10.67 400 1.2957 -11.1287 -12.9674 0.5700 1.8388 -239.8755 -221.9918 -1.1895 -1.2369
0.0001 13.33 500 1.3067 -11.1696 -13.0195 0.5700 1.8499 -240.3959 -222.4008 -1.1866 -1.2350
0.0001 16.0 600 1.3094 -11.2106 -13.0741 0.5700 1.8635 -240.9421 -222.8107 -1.1833 -1.2314
0.0001 18.67 700 1.3114 -11.2339 -13.0993 0.5700 1.8654 -241.1942 -223.0445 -1.1811 -1.2298
0.0001 21.33 800 1.3091 -11.2358 -13.1096 0.5700 1.8738 -241.2972 -223.0631 -1.1808 -1.2294
0.0001 24.0 900 1.3126 -11.2442 -13.1117 0.5700 1.8676 -241.3186 -223.1469 -1.1810 -1.2294
0.0001 26.67 1000 1.3096 -11.2358 -13.1040 0.5700 1.8682 -241.2410 -223.0633 -1.1809 -1.2295

Framework versions

  • PEFT 0.10.0
  • Transformers 4.39.3
  • Pytorch 2.2.2+cu121
  • Datasets 2.18.0
  • Tokenizers 0.15.2
Downloads last month
7
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for guoyu-zhang/model_usp4_dpo1

Adapter
(1208)
this model