---
library_name: transformers
base_model: qwen3-8b-base-sft-ultrachat-4xh200-batch-128-20260420-124036
tags:
- alignment-handbook
- epsilon-dpo
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: qwen3-8b-base-epsilon-dpo-ultrafeedback-4xh200-batch-128-20260420-124036
  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-epsilon-dpo-ultrafeedback-4xh200-batch-128-20260420-124036

This model is a fine-tuned version of [qwen3-8b-base-sft-ultrachat-4xh200-batch-128-20260420-124036](https://huggingface.co/qwen3-8b-base-sft-ultrachat-4xh200-batch-128-20260420-124036) on the HuggingFaceH4/ultrafeedback_binarized dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6403
- Epsilon Dpo/beta: 0.0021
- Epsilon Dpo/loss Margin Mean: 59.0314
- Epsilon Dpo/beta Margin Mean: 0.1219
- Epsilon Dpo/beta Margin Std: 0.2152
- Epsilon Dpo/beta Margin Grad Mean: -0.4699
- Epsilon Dpo/beta Margin Grad Std: 0.0531
- Rewards/chosen: -0.1383
- Rewards/rejected: -0.2601
- Rewards/accuracies: 0.7165
- Rewards/margins: 0.1219
- Logps/chosen: -346.2501
- Logps/rejected: -389.5577
- Logps/ref Chosen: -280.4283
- Logps/ref Rejected: -264.7045
- Logits/chosen: 1.5736
- Logits/rejected: 1.9569
- Kl/p Epsilon Steps: 0.7085
- Kl/n Epsilon Steps: 0.2855

## 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: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 8
- total_train_batch_size: 128
- total_eval_batch_size: 16
- 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 | Epsilon Dpo/beta | Epsilon Dpo/loss Margin Mean | Epsilon Dpo/beta Margin Mean | Epsilon Dpo/beta Margin Std | Epsilon Dpo/beta Margin Grad Mean | Epsilon Dpo/beta Margin Grad Std | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Logits/chosen | Logits/rejected | Kl/p Epsilon Steps | Kl/n Epsilon Steps |
|:-------------:|:------:|:----:|:---------------:|:----------------:|:----------------------------:|:----------------------------:|:---------------------------:|:---------------------------------:|:--------------------------------:|:--------------:|:----------------:|:------------------:|:---------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|:------------------:|:------------------:|
| 5.0674        | 0.4188 | 200  | 0.6322          | 0.0051           | 28.6770                      | 0.1452                       | 0.2575                      | -0.4644                           | 0.0631                           | -0.0590        | -0.2042          | 0.7170             | 0.1452          | -291.7776    | -304.7309      | -280.4283        | -264.7045          | 1.8063        | 2.1551          | 0.6990             | 0.2930             |
| 5.1073        | 0.8377 | 400  | 0.6403          | 0.0021           | 59.0314                      | 0.1219                       | 0.2152                      | -0.4699                           | 0.0531                           | -0.1383        | -0.2601          | 0.7165             | 0.1219          | -346.2501    | -389.5577      | -280.4283        | -264.7045          | 1.5736        | 1.9569          | 0.7085             | 0.2855             |


### Framework versions

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