---
library_name: transformers
base_model: jackf857/qwen3-8b-base-sft-hh-helpful-4xh200-batch-64-20260417-214452
tags:
- alignment-handbook
- margin-dpo
- generated_from_trainer
datasets:
- Anthropic/hh-rlhf
model-index:
- name: qwen3-8b-base-margin-dpo-hh-helpful-4xh200-batch-64-20260423-233948
  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-margin-dpo-hh-helpful-4xh200-batch-64-20260423-233948

This model is a fine-tuned version of [jackf857/qwen3-8b-base-sft-hh-helpful-4xh200-batch-64-20260417-214452](https://huggingface.co/jackf857/qwen3-8b-base-sft-hh-helpful-4xh200-batch-64-20260417-214452) on the Anthropic/hh-rlhf dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4195
- Margin Dpo/margin Mean: 15.8715
- Margin Dpo/margin Std: 17.0771
- Logps/chosen: -132.6461
- Logps/rejected: -139.3175
- Logps/ref Chosen: -101.8862
- Logps/ref Rejected: -92.6861
- Logits/chosen: -1.4538
- Logits/rejected: -1.1606

## 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: 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: 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 | Margin Dpo/margin Mean | Margin Dpo/margin Std | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Logits/chosen | Logits/rejected |
|:-------------:|:------:|:----:|:---------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|
| 1.0426        | 0.1468 | 100  | 0.6031          | 2.7546                 | 5.8844                | -105.6357    | -99.1902       | -101.8862        | -92.6861           | -0.0566       | 0.1974          |
| 0.7622        | 0.2937 | 200  | 0.4644          | 10.5567                | 12.6651               | -113.3752    | -114.7319      | -101.8862        | -92.6861           | -1.1210       | -0.8454         |
| 0.7508        | 0.4405 | 300  | 0.4348          | 13.3686                | 14.8396               | -121.8530    | -126.0214      | -101.8862        | -92.6861           | -1.2687       | -0.9791         |
| 0.4743        | 0.5874 | 400  | 0.4292          | 15.3820                | 16.7624               | -128.4094    | -134.5913      | -101.8862        | -92.6861           | -1.2117       | -0.8992         |
| 0.7107        | 0.7342 | 500  | 0.4213          | 15.8606                | 17.0950               | -131.5918    | -138.2523      | -101.8862        | -92.6861           | -1.3378       | -1.0359         |
| 0.5423        | 0.8811 | 600  | 0.4195          | 15.8715                | 17.0771               | -132.6461    | -139.3175      | -101.8862        | -92.6861           | -1.4538       | -1.1606         |


### Framework versions

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