|
--- |
|
base_model: alignment-handbook/zephyr-7b-sft-full |
|
datasets: |
|
- generation/UF6k |
|
library_name: peft |
|
license: apache-2.0 |
|
tags: |
|
- alignment-handbook |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
model-index: |
|
- name: zephyr-dpop-qlora-uf-oursuf6k-5e-7 |
|
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-dpop-qlora-uf-oursuf6k-5e-7 |
|
|
|
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 generation/UF6k dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.6903 |
|
- Positive Losses: 0.0369 |
|
- Dpo Losses: 0.6851 |
|
- Rewards/chosen: 0.0601 |
|
- Rewards/rejected: 0.0433 |
|
- Rewards/accuracies: 0.6090 |
|
- Rewards/margins: 0.0168 |
|
- Rewards/margins Max: 0.0933 |
|
- Rewards/margins Min: -0.0510 |
|
- Rewards/margins Std: 0.0482 |
|
- Logps/rejected: -254.2451 |
|
- Logps/chosen: -278.5801 |
|
- Logits/rejected: -2.7575 |
|
- Logits/chosen: -2.7964 |
|
|
|
## 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: 4 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 2 |
|
- gradient_accumulation_steps: 2 |
|
- total_train_batch_size: 16 |
|
- total_eval_batch_size: 16 |
|
- 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 | Positive Losses | Dpo Losses | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Rewards/margins Max | Rewards/margins Min | Rewards/margins Std | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | |
|
|:-------------:|:-----:|:----:|:---------------:|:---------------:|:----------:|:--------------:|:----------------:|:------------------:|:---------------:|:-------------------:|:-------------------:|:-------------------:|:--------------:|:------------:|:---------------:|:-------------:| |
|
| 0.6918 | 0.15 | 100 | 0.6930 | 0.0047 | 0.6922 | 0.0135 | 0.0117 | 0.6000 | 0.0019 | 0.0130 | -0.0081 | 0.0070 | -257.4135 | -283.2392 | -2.7718 | -2.8105 | |
|
| 0.6867 | 0.29 | 200 | 0.6917 | 0.0124 | 0.6900 | 0.0299 | 0.0235 | 0.6100 | 0.0063 | 0.0373 | -0.0214 | 0.0194 | -256.2250 | -281.6053 | -2.7690 | -2.8075 | |
|
| 0.6825 | 0.44 | 300 | 0.6909 | 0.0225 | 0.6879 | 0.0441 | 0.0333 | 0.625 | 0.0108 | 0.0619 | -0.0341 | 0.0319 | -255.2539 | -280.1855 | -2.7634 | -2.8021 | |
|
| 0.6808 | 0.58 | 400 | 0.6905 | 0.0314 | 0.6861 | 0.0543 | 0.0397 | 0.6170 | 0.0147 | 0.0808 | -0.0436 | 0.0415 | -254.6122 | -279.1606 | -2.7608 | -2.7994 | |
|
| 0.6769 | 0.73 | 500 | 0.6902 | 0.0330 | 0.6855 | 0.0585 | 0.0426 | 0.6120 | 0.0160 | 0.0885 | -0.0487 | 0.0458 | -254.3216 | -278.7400 | -2.7587 | -2.7975 | |
|
| 0.67 | 0.88 | 600 | 0.6903 | 0.0373 | 0.6851 | 0.0599 | 0.0432 | 0.6160 | 0.0167 | 0.0930 | -0.0511 | 0.0481 | -254.2623 | -278.6050 | -2.7602 | -2.7988 | |
|
|
|
|
|
### Framework versions |
|
|
|
- PEFT 0.7.1 |
|
- Transformers 4.39.0.dev0 |
|
- Pytorch 2.1.2+cu121 |
|
- Datasets 2.14.6 |
|
- Tokenizers 0.15.2 |
