---
library_name: transformers
license: apache-2.0
base_model: jackf857/qwen3-8b-base-sft-hh-helpful-4xh200-batch-64-20260417-214452
tags:
- alignment-handbook
- new-dpo
- generated_from_trainer
datasets:
- Anthropic/hh-rlhf
model-index:
- name: qwen3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.85
  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-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.85

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: 0.4843
- Fcm Dpo/beta: 0.0126
- Margin Dpo/margin Mean: 59.3348
- Margin Dpo/margin Std: 75.8064
- Logps/chosen: -239.6379
- Logps/rejected: -292.5469
- Logps/ref Chosen: -100.4936
- Logps/ref Rejected: -94.0678
- Logits/chosen: -1.6991
- Logits/rejected: -1.3195

## 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 | Fcm Dpo/beta | Margin Dpo/margin Mean | Margin Dpo/margin Std | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Logits/chosen | Logits/rejected |
|:-------------:|:------:|:----:|:---------------:|:------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|
| 0.8851        | 0.1468 | 100  | 0.5619          | 0.1911       | 2.7888                 | 5.1863                | -102.8106    | -99.1735       | -100.4936        | -94.0678           | -0.0489       | 0.1697          |
| 0.8588        | 0.2937 | 200  | 0.5018          | 0.0551       | 12.3878                | 17.0440               | -115.7780    | -121.7400      | -100.4936        | -94.0678           | -1.1454       | -0.8443         |
| 0.9115        | 0.4405 | 300  | 0.4888          | 0.0264       | 26.5166                | 33.8690               | -155.7471    | -175.8378      | -100.4936        | -94.0678           | -1.7435       | -1.4186         |
| 0.8025        | 0.5874 | 400  | 0.4957          | 0.0127       | 51.4539                | 65.7573               | -216.1685    | -261.1965      | -100.4936        | -94.0678           | -1.7851       | -1.4226         |
| 0.9588        | 0.7342 | 500  | 0.4848          | 0.0130       | 58.4689                | 75.4513               | -233.7258    | -285.7689      | -100.4936        | -94.0678           | -1.8270       | -1.4568         |
| 0.8671        | 0.8811 | 600  | 0.4843          | 0.0126       | 59.3348                | 75.8064               | -239.6379    | -292.5469      | -100.4936        | -94.0678           | -1.6991       | -1.3195         |


### Framework versions

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