Edit model card

zephyr-7b-dpo-qlora

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-qlora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5473
  • Rewards/chosen: -0.8609
  • Rewards/rejected: -1.5251
  • Rewards/accuracies: 0.7422
  • Rewards/margins: 0.6641
  • Logps/rejected: -404.3018
  • Logps/chosen: -336.2481
  • Logits/rejected: 0.0706
  • Logits/chosen: -0.1471

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-06
  • train_batch_size: 4
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 64
  • total_eval_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 1

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.6812 0.1 100 0.6787 0.0452 0.0120 0.6992 0.0332 -250.5929 -245.6322 -2.1942 -2.2517
0.6066 0.21 200 0.6151 -0.2303 -0.5020 0.6992 0.2717 -301.9975 -273.1855 -1.9906 -2.0610
0.5711 0.31 300 0.5927 -0.4441 -0.8513 0.7188 0.4072 -336.9228 -294.5666 -1.9417 -2.0223
0.557 0.42 400 0.5817 -0.5958 -1.0732 0.7227 0.4773 -359.1117 -309.7378 -1.7434 -1.8364
0.5703 0.52 500 0.5679 -0.7215 -1.2405 0.7266 0.5189 -375.8402 -322.3068 -0.8467 -0.9967
0.5498 0.63 600 0.5582 -0.7003 -1.2848 0.7578 0.5845 -380.2699 -320.1794 -0.2510 -0.4463
0.5279 0.73 700 0.5490 -0.8400 -1.4901 0.75 0.6501 -400.8082 -334.1553 0.0145 -0.1988
0.5264 0.84 800 0.5475 -0.8613 -1.5228 0.7461 0.6615 -404.0751 -336.2833 0.0604 -0.1549
0.5639 0.94 900 0.5475 -0.8628 -1.5267 0.7422 0.6639 -404.4688 -336.4348 0.0704 -0.1466

Framework versions

  • PEFT 0.7.1
  • Transformers 4.36.2
  • Pytorch 2.1.2+cu121
  • Datasets 2.14.6
  • Tokenizers 0.15.0
Downloads last month
64
Inference API
Unable to determine this model’s pipeline type. Check the docs .

Model tree for alignment-handbook/zephyr-7b-dpo-qlora

Adapter
(1170)
this model

Dataset used to train alignment-handbook/zephyr-7b-dpo-qlora

Collection including alignment-handbook/zephyr-7b-dpo-qlora