|
--- |
|
license: gemma |
|
base_model: tanliboy/zephyr-gemma-2-9b-sft |
|
tags: |
|
- alignment-handbook |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
datasets: |
|
- HuggingFaceH4/ultrafeedback_binarized |
|
model-index: |
|
- name: zephyr-gemma-2-9b-dpo-2 |
|
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. --> |
|
|
|
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/tanliboy/huggingface/runs/igf2ndcz) |
|
# zephyr-gemma-2-9b-dpo-2 |
|
|
|
This model is a fine-tuned version of [tanliboy/zephyr-gemma-2-9b-sft](https://huggingface.co/tanliboy/zephyr-gemma-2-9b-sft) on the HuggingFaceH4/ultrafeedback_binarized dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.5277 |
|
- Rewards/chosen: -0.6084 |
|
- Rewards/rejected: -1.2304 |
|
- Rewards/accuracies: 0.6880 |
|
- Rewards/margins: 0.6220 |
|
- Logps/rejected: -407.4499 |
|
- Logps/chosen: -375.1572 |
|
- Logits/rejected: -14.2928 |
|
- Logits/chosen: -14.1056 |
|
|
|
## 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: 2e-07 |
|
- train_batch_size: 2 |
|
- eval_batch_size: 2 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 8 |
|
- gradient_accumulation_steps: 4 |
|
- total_train_batch_size: 64 |
|
- 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 | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | |
|
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| |
|
| 0.6677 | 0.1047 | 100 | 0.6651 | 0.0269 | -0.0301 | 0.6440 | 0.0570 | -287.4193 | -311.6301 | -9.5213 | -9.2788 | |
|
| 0.5915 | 0.2094 | 200 | 0.5920 | -0.3361 | -0.6472 | 0.6880 | 0.3110 | -349.1276 | -347.9349 | -11.9562 | -11.6561 | |
|
| 0.5723 | 0.3141 | 300 | 0.5674 | -0.3955 | -0.7898 | 0.6880 | 0.3943 | -363.3917 | -353.8749 | -12.6873 | -12.4526 | |
|
| 0.5622 | 0.4187 | 400 | 0.5468 | -0.5688 | -1.0827 | 0.6800 | 0.5139 | -392.6759 | -371.2007 | -14.2367 | -13.9401 | |
|
| 0.5441 | 0.5234 | 500 | 0.5363 | -0.6274 | -1.2091 | 0.6680 | 0.5817 | -405.3189 | -377.0607 | -14.3976 | -14.1308 | |
|
| 0.5125 | 0.6281 | 600 | 0.5344 | -0.5757 | -1.1705 | 0.6840 | 0.5948 | -401.4605 | -371.8937 | -14.3713 | -14.1120 | |
|
| 0.5158 | 0.7328 | 700 | 0.5316 | -0.6220 | -1.2328 | 0.6760 | 0.6108 | -407.6867 | -376.5182 | -14.2832 | -14.1010 | |
|
| 0.5133 | 0.8375 | 800 | 0.5278 | -0.6258 | -1.2452 | 0.6800 | 0.6193 | -408.9254 | -376.9043 | -14.2747 | -14.0908 | |
|
| 0.5098 | 0.9422 | 900 | 0.5276 | -0.6043 | -1.2270 | 0.6960 | 0.6227 | -407.1073 | -374.7531 | -14.2849 | -14.1010 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.43.1 |
|
- Pytorch 2.3.1+cu121 |
|
- Datasets 2.19.1 |
|
- Tokenizers 0.19.1 |
|
|
