|
|
--- |
|
|
library_name: transformers |
|
|
license: apache-2.0 |
|
|
base_model: alignment-handbook/zephyr-7b-sft-full |
|
|
tags: |
|
|
- alignment-handbook |
|
|
- trl |
|
|
- dpo |
|
|
- generated_from_trainer |
|
|
- trl |
|
|
- dpo |
|
|
- generated_from_trainer |
|
|
datasets: |
|
|
- HuggingFaceH4/ultrafeedback_binarized |
|
|
model-index: |
|
|
- name: zephyr-7b-align-scan-3e-07-0.62-polynomial-3.0 |
|
|
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-align-scan-3e-07-0.62-polynomial-3.0 |
|
|
|
|
|
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 dataset. |
|
|
It achieves the following results on the evaluation set: |
|
|
- Loss: 0.8726 |
|
|
- Rewards/chosen: -0.1874 |
|
|
- Rewards/rejected: -1.8304 |
|
|
- Rewards/accuracies: 0.375 |
|
|
- Rewards/margins: 1.6430 |
|
|
- Logps/rejected: -84.0806 |
|
|
- Logps/chosen: -74.7935 |
|
|
- Logits/rejected: -2.6285 |
|
|
- Logits/chosen: -2.6453 |
|
|
|
|
|
## 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: 3e-07 |
|
|
- train_batch_size: 8 |
|
|
- eval_batch_size: 8 |
|
|
- seed: 42 |
|
|
- distributed_type: multi-GPU |
|
|
- num_devices: 4 |
|
|
- gradient_accumulation_steps: 2 |
|
|
- total_train_batch_size: 64 |
|
|
- total_eval_batch_size: 32 |
|
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
|
- lr_scheduler_type: polynomial |
|
|
- 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 | |
|
|
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| |
|
|
| 0.6629 | 0.3484 | 100 | 0.6374 | 0.7586 | 0.3997 | 0.3452 | 0.3589 | -80.4837 | -73.2678 | -2.5455 | -2.5615 | |
|
|
| 0.7044 | 0.6969 | 200 | 0.6785 | 0.6115 | 0.1187 | 0.3353 | 0.4927 | -80.9369 | -73.5050 | -2.5325 | -2.5487 | |
|
|
| 0.3945 | 1.0453 | 300 | 0.6975 | 0.7667 | 0.1071 | 0.3552 | 0.6597 | -80.9557 | -73.2546 | -2.5596 | -2.5753 | |
|
|
| 0.3859 | 1.3937 | 400 | 0.7396 | 1.4671 | 0.5658 | 0.3571 | 0.9013 | -80.2158 | -72.1250 | -2.5834 | -2.5995 | |
|
|
| 0.3893 | 1.7422 | 500 | 0.7904 | -0.4771 | -1.4060 | 0.3492 | 0.9290 | -83.3962 | -75.2607 | -2.6499 | -2.6659 | |
|
|
| 0.3749 | 2.0906 | 600 | 0.8125 | 0.5611 | -0.4847 | 0.3631 | 1.0458 | -81.9100 | -73.5862 | -2.6159 | -2.6321 | |
|
|
| 0.3662 | 2.4390 | 700 | 0.8412 | -0.6104 | -2.0869 | 0.3651 | 1.4765 | -84.4944 | -75.4757 | -2.5941 | -2.6112 | |
|
|
| 0.3615 | 2.7875 | 800 | 0.8766 | -0.9523 | -2.5666 | 0.3611 | 1.6143 | -85.2680 | -76.0272 | -2.6367 | -2.6538 | |
|
|
|
|
|
|
|
|
### Framework versions |
|
|
|
|
|
- Transformers 4.44.2 |
|
|
- Pytorch 2.4.0 |
|
|
- Datasets 2.21.0 |
|
|
- Tokenizers 0.19.1 |
|
|
|
