model_usp2_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: 1.0445
  • Rewards/chosen: -10.7928
  • Rewards/rejected: -14.2959
  • Rewards/accuracies: 0.7400
  • Rewards/margins: 3.5031
  • Logps/rejected: -250.8612
  • Logps/chosen: -215.1325
  • Logits/rejected: -0.8799
  • Logits/chosen: -0.9353

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
0.0632 2.67 100 0.6858 -5.5164 -7.1903 0.6900 1.6739 -179.8054 -162.3683 -0.8170 -0.7196
0.002 5.33 200 0.7615 -6.7178 -9.5839 0.7600 2.8661 -203.7411 -174.3826 -1.0768 -1.0701
0.0001 8.0 300 1.0247 -10.4976 -13.8923 0.7400 3.3948 -246.8256 -212.1801 -0.8995 -0.9506
0.0001 10.67 400 1.0323 -10.6255 -14.0760 0.75 3.4505 -248.6621 -213.4589 -0.8910 -0.9437
0.0001 13.33 500 1.0328 -10.7107 -14.1992 0.7400 3.4885 -249.8943 -214.3115 -0.8858 -0.9397
0.0001 16.0 600 1.0378 -10.7577 -14.2607 0.7400 3.5030 -250.5091 -214.7812 -0.8823 -0.9372
0.0 18.67 700 1.0407 -10.7811 -14.2886 0.75 3.5075 -250.7885 -215.0155 -0.8811 -0.9363
0.0001 21.33 800 1.0415 -10.7857 -14.2997 0.7400 3.5139 -250.8989 -215.0617 -0.8802 -0.9359
0.0001 24.0 900 1.0423 -10.7886 -14.2954 0.7400 3.5068 -250.8562 -215.0906 -0.8802 -0.9356
0.0001 26.67 1000 1.0445 -10.7928 -14.2959 0.7400 3.5031 -250.8612 -215.1325 -0.8799 -0.9353

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
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for guoyu-zhang/model_usp2_dpo1

Adapter
(1202)
this model