model_usp2_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: 3.0422
  • Rewards/chosen: -16.6521
  • Rewards/rejected: -22.4980
  • Rewards/accuracies: 0.6900
  • Rewards/margins: 5.8460
  • Logps/rejected: -134.2057
  • Logps/chosen: -127.2480
  • Logits/rejected: -0.3729
  • Logits/chosen: -0.3027

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
0.1118 2.67 100 1.5049 0.5063 -2.6419 0.6100 3.1482 -112.1433 -108.1832 -0.0840 -0.0663
0.0485 5.33 200 3.9547 -16.7119 -24.8420 0.6600 8.1301 -136.8102 -127.3146 -0.4936 -0.4252
0.0933 8.0 300 2.7780 6.4777 2.8373 0.6200 3.6404 -106.0553 -101.5483 0.0824 0.1237
0.0001 10.67 400 3.9997 -24.6430 -30.7416 0.6600 6.0986 -143.3652 -136.1268 -0.5838 -0.5341
0.0 13.33 500 3.0680 -16.6486 -22.4522 0.6900 5.8036 -134.1547 -127.2442 -0.3725 -0.3021
0.0 16.0 600 3.0371 -16.6460 -22.4827 0.6900 5.8367 -134.1887 -127.2413 -0.3725 -0.3022
0.0 18.67 700 3.0540 -16.6424 -22.4815 0.6900 5.8391 -134.1873 -127.2373 -0.3728 -0.3028
0.0 21.33 800 3.0298 -16.6187 -22.4938 0.6900 5.8750 -134.2010 -127.2110 -0.3731 -0.3028
0.0 24.0 900 3.0554 -16.6241 -22.4802 0.6900 5.8561 -134.1858 -127.2169 -0.3725 -0.3027
0.0 26.67 1000 3.0422 -16.6521 -22.4980 0.6900 5.8460 -134.2057 -127.2480 -0.3729 -0.3027

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_dpo9

Adapter
(1202)
this model