---
library_name: transformers
base_model: jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128
tags:
- alignment-handbook
- orpo
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: qwen3-8b-base-orpo-ultrafeedback-4xh200-batch-128
  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. -->

# qwen3-8b-base-orpo-ultrafeedback-4xh200-batch-128

This model is a fine-tuned version of [jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128](https://huggingface.co/jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128) on the HuggingFaceH4/ultrafeedback_binarized dataset.
It achieves the following results on the evaluation set:
- Loss: 1.0784
- Rewards/chosen: -0.0085
- Rewards/rejected: -0.0105
- Rewards/accuracies: 0.6060
- Rewards/margins: 0.0020
- Logps/rejected: -1.0496
- Logps/chosen: -0.8523
- Logits/rejected: 2.1915
- Logits/chosen: 2.1569
- Nll Loss: 1.1100
- Log Odds Ratio: -0.6630
- Log Odds Chosen: 0.3031

## 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: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 8
- total_train_batch_size: 128
- total_eval_batch_size: 8
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- 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 | Nll Loss | Log Odds Ratio | Log Odds Chosen |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------:|:--------------:|:---------------:|
| 8.6911        | 0.4188 | 200  | 1.0935          | -0.0086        | -0.0106          | 0.6100             | 0.0020          | -1.0645        | -0.8642      | 2.0572          | 2.0427        | 1.1233   | -0.6611        | 0.3058          |
| 8.6763        | 0.8377 | 400  | 1.0784          | -0.0085        | -0.0105          | 0.6060             | 0.0020          | -1.0496        | -0.8523      | 2.1915          | 2.1569        | 1.1100   | -0.6630        | 0.3031          |


### Framework versions

- Transformers 4.51.0
- Pytorch 2.3.1+cu121
- Datasets 2.21.0
- Tokenizers 0.21.4