---
library_name: transformers
base_model: timarni/qwen3_dpo
tags:
- generated_from_trainer
datasets:
- timarni/MNLP_intstruction_tuning
model-index:
- name: outputs/dpo_full_alpaca
  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. -->

[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.9.2`
```yaml
base_model: timarni/qwen3_dpo
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
strict: false

chat_template: qwen3
datasets:
  - path: timarni/MNLP_intstruction_tuning
    type: alpaca
    split: train

shuffle_merged_datasets: true

val_set_size: 0.1
output_dir: ./outputs/dpo_full_alpaca
dataset_prepared_path: last_run_prepared

sequence_len: 4096 #2048
sample_packing: true # was true -> need to check if it actually learns on the samples or not (better understand te hyperparam and event. install axolotl to debug)
eval_sample_packing: true
pad_to_sequence_len: true
# train_on_inputs: true # NEW
# group_by_length: false NEW?

# To be sure that no LORA is done
adapter: null
lora: false
merge_lora: false

wandb_project: mnlp_project
wandb_entity: tim-arni
wandb_watch:
wandb_name: dpo_full_alpaca_resume_from_ckpt
wandb_log_model:

gradient_accumulation_steps: 16 # 2
micro_batch_size: 2 # 1
num_epochs: 3
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.00005 # 0.00005
# cosine_min_lr_ratio: 0.1

warmup_ratio: 0.05
weight_decay: 0.01

bf16: auto
tf32: true

gradient_checkpointing: offload
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint: /mloscratch/users/arni/Workspace/mnlp_sft/outputs/dpo_full_alpaca/checkpoint-186
logging_steps: 1
gradient_clipping: 1.0 # or max_grad_norm?
flash_attention: true

evals_per_epoch: 2
saves_per_epoch: 1
save_total_limit: 20
special_tokens:

```

</details><br>

# outputs/dpo_full_alpaca

This model is a fine-tuned version of [timarni/qwen3_dpo](https://huggingface.co/timarni/qwen3_dpo) on the timarni/MNLP_intstruction_tuning dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1520

## 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-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 16
- total_train_batch_size: 128
- total_eval_batch_size: 8
- optimizer: Use 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_steps: 13
- num_epochs: 3.0

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 0.7154        | 0.0107 | 1    | 1.1239          |
| 0.1282        | 0.2567 | 24   | 0.2029          |
| 0.1105        | 0.5134 | 48   | 0.1860          |
| 0.1056        | 0.7701 | 72   | 0.1779          |
| 0.1004        | 1.0214 | 96   | 0.1736          |
| 0.0912        | 1.2781 | 120  | 0.1643          |
| 0.0861        | 1.5348 | 144  | 0.1576          |
| 0.0791        | 1.7914 | 168  | 0.1530          |
| 0.0751        | 2.0642 | 192  | 0.1510          |
| 0.0625        | 2.3209 | 216  | 0.1509          |
| 0.0453        | 2.5775 | 240  | 0.1513          |
| 0.0426        | 2.8342 | 264  | 0.1520          |


### Framework versions

- Transformers 4.51.3
- Pytorch 2.5.1+cu121
- Datasets 3.5.1
- Tokenizers 0.21.1