---
library_name: transformers
base_model: jackf857/qwen3-8b-base-sft-hh-harmless-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-harmless-4xh200-batch-64-20260423-234249
  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-harmless-4xh200-batch-64-20260423-234249

This model is a fine-tuned version of [jackf857/qwen3-8b-base-sft-hh-harmless-4xh200-batch-64-20260417-214452](https://huggingface.co/jackf857/qwen3-8b-base-sft-hh-harmless-4xh200-batch-64-20260417-214452) on the Anthropic/hh-rlhf dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5180
- Margin Dpo/margin Mean: 7.8948
- Margin Dpo/margin Std: 11.6820
- Logps/chosen: -90.0938
- Logps/rejected: -105.9037
- Logps/ref Chosen: -87.3172
- Logps/ref Rejected: -95.2323
- Logits/chosen: 1.4433
- Logits/rejected: 1.3188

## 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 |
|:-------------:|:------:|:----:|:---------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|
| 1.3236        | 0.1512 | 100  | 0.6540          | 0.9206                 | 1.8427                | -86.6428     | -95.4785       | -87.3172         | -95.2323           | 1.6973        | 1.5878          |
| 1.1498        | 0.3023 | 200  | 0.5566          | 5.3407                 | 9.0153                | -88.1022     | -101.3580      | -87.3172         | -95.2323           | 1.4121        | 1.2978          |
| 1.1522        | 0.4535 | 300  | 0.5328          | 7.2941                 | 11.5055               | -91.8542     | -107.0635      | -87.3172         | -95.2323           | 1.4997        | 1.3738          |
| 1.2091        | 0.6047 | 400  | 0.5248          | 7.2854                 | 11.1368               | -89.2882     | -104.4887      | -87.3172         | -95.2323           | 1.4582        | 1.3356          |
| 1.0214        | 0.7559 | 500  | 0.5192          | 8.0772                 | 11.9903               | -90.4015     | -106.3938      | -87.3172         | -95.2323           | 1.7114        | 1.5744          |
| 1.1318        | 0.9070 | 600  | 0.5180          | 7.8948                 | 11.6820               | -90.0938     | -105.9037      | -87.3172         | -95.2323           | 1.4433        | 1.3188          |


### Framework versions

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