Hugging Face's logo

wzhouad
/
zephyr-7b-dpo-full

Text Generation
Transformers
Safetensors
llama
trl
dpo
Generated from Trainer
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
zephyr-7b-dpo-full / README.md
wzhouad's picture
wzhouad
Model save
fe08d08 verified
preview code
|
raw
history blame
No virus
5.36 kB
metadata
license: mit
base_model: HuggingFaceH4/mistral-7b-sft-beta
tags:
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: zephyr-7b-dpo-full
    results: []

Visualize in Weights & Biases

zephyr-7b-dpo-full

This model is a fine-tuned version of HuggingFaceH4/mistral-7b-sft-beta on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0227
  • Rewards/chosen: -2.3113
  • Rewards/rejected: -2.8479
  • Rewards/accuracies: 0.6931
  • Rewards/margins: 0.5365
  • Logps/rejected: -435.4867
  • Logps/chosen: -375.3782
  • Logits/rejected: -1.4622
  • Logits/chosen: -1.5834
  • Debug/policy Weights: 0.0374
  • Debug/losses: 0.0212
  • Debug/raw Losses: 0.5682

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-07
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 128
  • 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 Debug/policy Weights Debug/losses Debug/raw Losses
0.1734 0.0796 100 0.1631 -0.1425 -0.1765 0.5924 0.0340 -168.3475 -158.4907 -2.7043 -2.7124 0.2381 0.1616 0.6787
0.0795 0.1592 200 0.0826 -0.7160 -0.9309 0.6483 0.2150 -243.7922 -215.8411 -2.4879 -2.4997 0.1266 0.0800 0.6296
0.0545 0.2388 300 0.0572 -1.0974 -1.4187 0.6642 0.3213 -292.5661 -253.9808 -2.4160 -2.4302 0.0894 0.0550 0.6166
0.0288 0.3183 400 0.0302 -1.9563 -2.3772 0.6698 0.4209 -388.4184 -339.8692 -2.2376 -2.2573 0.0477 0.0287 0.6044
0.0358 0.3979 500 0.0407 -1.7169 -2.1543 0.6698 0.4374 -366.1241 -315.9322 -2.2265 -2.2540 0.0659 0.0394 0.6064
0.0309 0.4775 600 0.0302 -1.9504 -2.4092 0.6660 0.4587 -391.6147 -339.2857 -2.0849 -2.1159 0.0489 0.0287 0.5899
0.0203 0.5571 700 0.0198 -2.3315 -2.7643 0.6856 0.4328 -427.1261 -377.3937 -1.6613 -1.7384 0.0317 0.0185 0.5808
0.0192 0.6367 800 0.0182 -2.5929 -3.1225 0.6866 0.5297 -462.9526 -403.5321 -1.0483 -1.2122 0.0290 0.0169 0.5789
0.0233 0.7163 900 0.0237 -2.3310 -2.8931 0.6810 0.5621 -440.0111 -377.3470 -1.3096 -1.4493 0.0387 0.0221 0.5726
0.0213 0.7959 1000 0.0219 -2.4229 -2.9606 0.6931 0.5377 -446.7564 -386.5316 -1.4880 -1.6049 0.0357 0.0203 0.5694
0.0229 0.8754 1100 0.0231 -2.2736 -2.7873 0.6950 0.5137 -429.4283 -371.6010 -1.5527 -1.6574 0.0379 0.0215 0.5695
0.0216 0.9550 1200 0.0227 -2.3113 -2.8479 0.6931 0.5365 -435.4867 -375.3782 -1.4622 -1.5834 0.0374 0.0212 0.5682

Framework versions

  • Transformers 4.41.0.dev0
  • Pytorch 2.1.2+cu121
  • Datasets 2.14.6
  • Tokenizers 0.19.1