|
--- |
|
license: apache-2.0 |
|
base_model: mistralai/Mistral-7B-v0.1 |
|
tags: |
|
- trl |
|
- orpo |
|
- alignment-handbook |
|
- generated_from_trainer |
|
model-index: |
|
- name: zephyr-7b-sft-full-orpo |
|
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/statking/huggingface/runs/ehjj41t1) |
|
# zephyr-7b-sft-full-orpo |
|
|
|
This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on an unknown dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.5088 |
|
- Rewards/chosen: -0.0404 |
|
- Rewards/rejected: -0.0510 |
|
- Rewards/accuracies: 0.6290 |
|
- Rewards/margins: 0.0106 |
|
- Logps/rejected: -1.0202 |
|
- Logps/chosen: -0.8085 |
|
- Logits/rejected: -2.5337 |
|
- Logits/chosen: -2.5634 |
|
- Nll Loss: 0.4741 |
|
- Log Odds Ratio: -0.6379 |
|
- Log Odds Chosen: 0.3305 |
|
|
|
## 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-05 |
|
- 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: inverse_sqrt |
|
- lr_scheduler_warmup_steps: 100 |
|
- 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 | Nll Loss | Log Odds Ratio | Log Odds Chosen | |
|
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------:|:--------------:|:---------------:| |
|
| 0.5707 | 0.1049 | 100 | 1.1268 | -0.0452 | -0.0539 | 0.6369 | 0.0086 | -1.0774 | -0.9045 | -2.5432 | -2.5811 | 1.0893 | -0.6413 | 0.2601 | |
|
| 0.5663 | 0.2098 | 200 | 0.5741 | -0.0440 | -0.0534 | 0.6270 | 0.0094 | -1.0676 | -0.8799 | -2.5377 | -2.5597 | 0.5352 | -0.6447 | 0.2863 | |
|
| 0.5817 | 0.3146 | 300 | 0.5572 | -0.0440 | -0.0531 | 0.6190 | 0.0091 | -1.0628 | -0.8808 | -2.4499 | -2.4818 | 0.5207 | -0.6503 | 0.2780 | |
|
| 0.5724 | 0.4195 | 400 | 0.5416 | -0.0426 | -0.0515 | 0.625 | 0.0089 | -1.0293 | -0.8510 | -2.4026 | -2.4376 | 0.5060 | -0.6551 | 0.2819 | |
|
| 0.5486 | 0.5244 | 500 | 0.5344 | -0.0425 | -0.0526 | 0.6151 | 0.0101 | -1.0514 | -0.8492 | -2.4373 | -2.4718 | 0.4990 | -0.6439 | 0.3193 | |
|
| 0.5156 | 0.6293 | 600 | 0.5242 | -0.0417 | -0.0514 | 0.6151 | 0.0098 | -1.0285 | -0.8333 | -2.5551 | -2.5811 | 0.4882 | -0.6470 | 0.3056 | |
|
| 0.5297 | 0.7341 | 700 | 0.5191 | -0.0411 | -0.0521 | 0.6310 | 0.0110 | -1.0422 | -0.8215 | -2.4477 | -2.4801 | 0.4838 | -0.6351 | 0.3407 | |
|
| 0.5184 | 0.8390 | 800 | 0.5138 | -0.0409 | -0.0532 | 0.6310 | 0.0123 | -1.0647 | -0.8179 | -2.4575 | -2.4922 | 0.4796 | -0.6304 | 0.3783 | |
|
| 0.5235 | 0.9439 | 900 | 0.5088 | -0.0404 | -0.0510 | 0.6290 | 0.0106 | -1.0202 | -0.8085 | -2.5337 | -2.5634 | 0.4741 | -0.6379 | 0.3305 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.41.0.dev0 |
|
- Pytorch 2.3.0+cu121 |
|
- Datasets 2.19.1 |
|
- Tokenizers 0.19.1 |
|
|
