zephyr-7b-dpo-full / README.md
lewtun's picture
lewtun HF staff
Update README.md
2eb22f9
|
raw
history blame
7.62 kB
---
license: apache-2.0
base_model: alignment-handbook/zephyr-7b-sft-full
tags:
- generated_from_trainer
- alignment-handbook
model-index:
- name: zephyr-7b-dpo-full
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# zephyr-7b-dpo-full
This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the [HuggingFaceH4/ultrafeedback_binarized](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized) dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6910
- Rewards/chosen: -3.9218
- Rewards/rejected: -8.2942
- Rewards/accuracies: 0.8125
- Rewards/margins: 4.3724
- Logps/rejected: -279.5480
- Logps/chosen: -293.9998
- Logits/rejected: -2.6725
- Logits/chosen: -2.7826
## 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: 2
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 32
- total_train_batch_size: 64
- total_eval_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.5386 | 0.1 | 100 | 0.5208 | 0.0564 | -0.7521 | 0.7188 | 0.8085 | -204.1269 | -254.2179 | -3.0136 | -3.0550 |
| 0.4931 | 0.21 | 200 | 0.4882 | -0.0132 | -1.2683 | 0.7812 | 1.2551 | -209.2889 | -254.9136 | -3.1056 | -3.1407 |
| 0.479 | 0.31 | 300 | 0.5038 | -0.1035 | -1.4012 | 0.7812 | 1.2978 | -210.6186 | -255.8163 | -3.0809 | -3.1328 |
| 0.5052 | 0.41 | 400 | 0.5154 | -0.1923 | -1.8783 | 0.7969 | 1.6860 | -215.3891 | -256.7043 | -2.9104 | -2.9644 |
| 0.4513 | 0.52 | 500 | 0.4979 | 0.0207 | -1.6562 | 0.7969 | 1.6769 | -213.1682 | -254.5742 | -3.0061 | -3.0657 |
| 0.4905 | 0.62 | 600 | 0.4907 | -0.0944 | -1.5847 | 0.7656 | 1.4903 | -212.4527 | -255.7256 | -2.9374 | -3.0170 |
| 0.5609 | 0.72 | 700 | 0.4928 | -0.4249 | -1.7238 | 0.7656 | 1.2989 | -213.8441 | -259.0304 | -2.9475 | -3.0128 |
| 0.5338 | 0.83 | 800 | 0.4767 | -0.1567 | -1.9114 | 0.8125 | 1.7547 | -215.7200 | -256.3484 | -2.8455 | -2.9183 |
| 0.5039 | 0.93 | 900 | 0.4854 | -0.0886 | -1.6900 | 0.75 | 1.6014 | -213.5057 | -255.6674 | -2.8295 | -2.9093 |
| 0.0776 | 1.03 | 1000 | 0.4938 | -0.4848 | -2.5287 | 0.7656 | 2.0438 | -221.8927 | -259.6300 | -2.7580 | -2.8437 |
| 0.0901 | 1.14 | 1100 | 0.5071 | -1.0800 | -3.2419 | 0.7812 | 2.1619 | -229.0247 | -265.5817 | -2.8036 | -2.8858 |
| 0.0828 | 1.24 | 1200 | 0.5159 | -0.9682 | -3.4087 | 0.7812 | 2.4406 | -230.6935 | -264.4635 | -2.7961 | -2.8708 |
| 0.0916 | 1.34 | 1300 | 0.5222 | -1.0832 | -3.5535 | 0.7969 | 2.4703 | -232.1411 | -265.6135 | -2.8019 | -2.8754 |
| 0.0965 | 1.44 | 1400 | 0.5204 | -1.1951 | -3.5681 | 0.7969 | 2.3731 | -232.2874 | -266.7324 | -2.8058 | -2.8884 |
| 0.0716 | 1.55 | 1500 | 0.5381 | -1.6588 | -4.0838 | 0.7188 | 2.4250 | -237.4441 | -271.3697 | -2.7979 | -2.8862 |
| 0.0957 | 1.65 | 1600 | 0.5151 | -1.1746 | -3.7477 | 0.75 | 2.5731 | -234.0834 | -266.5278 | -2.7960 | -2.8976 |
| 0.0645 | 1.75 | 1700 | 0.5393 | -1.7591 | -4.6011 | 0.8125 | 2.8419 | -242.6167 | -272.3728 | -2.7483 | -2.8592 |
| 0.0838 | 1.86 | 1800 | 0.5385 | -1.6606 | -4.4648 | 0.7656 | 2.8042 | -241.2545 | -271.3875 | -2.7311 | -2.8383 |
| 0.1106 | 1.96 | 1900 | 0.5322 | -1.5621 | -3.9779 | 0.7969 | 2.4158 | -236.3850 | -270.4025 | -2.8194 | -2.9133 |
| 0.0174 | 2.06 | 2000 | 0.5921 | -2.4968 | -5.9514 | 0.7969 | 3.4546 | -256.1199 | -279.7498 | -2.7579 | -2.8631 |
| 0.0134 | 2.17 | 2100 | 0.6247 | -2.9002 | -6.4277 | 0.7969 | 3.5275 | -260.8829 | -283.7838 | -2.7316 | -2.8319 |
| 0.0148 | 2.27 | 2200 | 0.6402 | -3.2520 | -7.0627 | 0.7812 | 3.8106 | -267.2330 | -287.3020 | -2.6991 | -2.8064 |
| 0.0142 | 2.37 | 2300 | 0.6563 | -3.2715 | -7.1303 | 0.8281 | 3.8588 | -267.9088 | -287.4962 | -2.6871 | -2.7992 |
| 0.011 | 2.48 | 2400 | 0.6605 | -3.2996 | -7.2258 | 0.7969 | 3.9262 | -268.8643 | -287.7776 | -2.6555 | -2.7717 |
| 0.0065 | 2.58 | 2500 | 0.6935 | -3.6399 | -8.0232 | 0.8125 | 4.3832 | -276.8377 | -291.1808 | -2.6780 | -2.7902 |
| 0.0089 | 2.68 | 2600 | 0.6773 | -3.4822 | -7.8182 | 0.8125 | 4.3360 | -274.7881 | -289.6033 | -2.6885 | -2.7994 |
| 0.0102 | 2.79 | 2700 | 0.6813 | -3.5909 | -7.8097 | 0.8281 | 4.2187 | -274.7028 | -290.6908 | -2.6877 | -2.7970 |
| 0.0136 | 2.89 | 2800 | 0.6892 | -3.8236 | -8.1490 | 0.8125 | 4.3254 | -278.0957 | -293.0175 | -2.6765 | -2.7862 |
| 0.0091 | 2.99 | 2900 | 0.6913 | -3.9199 | -8.3004 | 0.8125 | 4.3806 | -279.6104 | -293.9802 | -2.6728 | -2.7830 |
### Framework versions
- Transformers 4.35.0
- Pytorch 2.1.0+cu118
- Datasets 2.14.6
- Tokenizers 0.14.1