---
library_name: transformers
base_model: jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128
tags:
- alignment-handbook
- new-dpo
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725
  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-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725

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: 0.5897
- Fcm Dpo/beta: 0.0056
- Margin Dpo/margin Mean: 51.0513
- Margin Dpo/margin Std: 81.3439
- Logps/chosen: -321.9651
- Logps/rejected: -357.2452
- Logps/ref Chosen: -280.4167
- Logps/ref Rejected: -264.6455
- Kl/chosen Kl Mean: -41.5484
- Kl/rejected Kl Mean: -92.5997
- Kl/mean: -67.0741
- Kl/std: 70.1467
- Logits/chosen: 1.5636
- Logits/rejected: 1.6451

## 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 | Kl/chosen Kl Mean | Kl/mean  | Kl/rejected Kl Mean | Kl/std  | Fcm Dpo/beta | Logits/chosen | Logits/rejected | Logps/chosen | Logps/ref Chosen | Logps/ref Rejected | Logps/rejected | Validation Loss | Margin Dpo/margin Mean | Margin Dpo/margin Std |
|:-------------:|:------:|:----:|:-----------------:|:--------:|:-------------------:|:-------:|:------------:|:-------------:|:---------------:|:------------:|:----------------:|:------------------:|:--------------:|:---------------:|:----------------------:|:---------------------:|
| 4.8732        | 0.4188 | 200  | -7.6098           | -20.6141 | -33.6184            | 36.4948 | 0.0114       | 1.5705        | 1.6149          | -288.0265    | -280.4167        | -264.6455          | -298.2639      | 0.5900          | 26.0086                | 42.7299               |
| 4.6016        | 0.8377 | 400  | 0.5897            | 0.0056   | 51.0513             | 81.3439 | -321.9651    | -357.2452     | -280.4167       | -264.6455    | -41.5484         | -92.5997           | -67.0741       | 70.1467         | 1.5636                 | 1.6451                |


### Framework versions

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