model_usp4_dpo9

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: 3.0245
  • Rewards/chosen: -8.7593
  • Rewards/rejected: -14.6879
  • Rewards/accuracies: 0.6800
  • Rewards/margins: 5.9286
  • Logps/rejected: -126.1702
  • Logps/chosen: -122.7822
  • Logits/rejected: -0.5270
  • Logits/chosen: -0.4937

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
0.0474 2.67 100 1.4973 -7.5502 -10.3869 0.6800 2.8366 -121.3912 -121.4387 -0.3436 -0.3419
0.0519 5.33 200 2.4820 -8.4480 -12.1331 0.6800 3.6850 -123.3315 -122.4363 -0.4441 -0.4246
0.042 8.0 300 2.9082 -4.2282 -9.0990 0.6700 4.8708 -119.9602 -117.7476 -0.3275 -0.2971
0.0 10.67 400 3.0274 -8.7246 -14.6605 0.6900 5.9359 -126.1397 -122.7436 -0.5266 -0.4935
0.0 13.33 500 3.0135 -8.7321 -14.6991 0.6900 5.9670 -126.1826 -122.7520 -0.5276 -0.4944
0.0 16.0 600 3.0025 -8.7128 -14.6671 0.7000 5.9542 -126.1470 -122.7305 -0.5289 -0.4956
0.0 18.67 700 3.0086 -8.7343 -14.6314 0.6900 5.8971 -126.1074 -122.7544 -0.5275 -0.4944
0.0 21.33 800 3.0003 -8.7154 -14.6723 0.6800 5.9569 -126.1529 -122.7334 -0.5277 -0.4945
0.0 24.0 900 3.0039 -8.7302 -14.6820 0.6900 5.9518 -126.1636 -122.7498 -0.5274 -0.4944
0.0 26.67 1000 3.0245 -8.7593 -14.6879 0.6800 5.9286 -126.1702 -122.7822 -0.5270 -0.4937

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
3
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for guoyu-zhang/model_usp4_dpo9

Adapter
(1208)
this model