---
library_name: peft
license: mit
base_model: deepseek-ai/Deepseek-R1-Distill-Qwen-32B
tags:
- alignment-handbook
- trl
- sft
- generated_from_trainer
datasets:
- tttx/r1-trajectories-collection-round-2
- tttx/r1-trajectories-arcagi-barc
model-index:
- name: qwen-32b-sft-full-arcagi-barc
  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. -->

# qwen-32b-sft-full-arcagi-barc

This model is a fine-tuned version of [deepseek-ai/Deepseek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/Deepseek-R1-Distill-Qwen-32B) on the tttx/r1-trajectories-collection-round-2 and the tttx/r1-trajectories-arcagi-barc datasets.
It achieves the following results on the evaluation set:
- Loss: 0.4392

## 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: 4
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 32
- 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_ratio: 0.1
- num_epochs: 4

### Training results

| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 0.4064        | 1.0   | 328  | 0.4612          |
| 0.3589        | 2.0   | 656  | 0.4404          |
| 0.3656        | 3.0   | 984  | 0.4399          |
| 0.3443        | 4.0   | 1312 | 0.4392          |


### Framework versions

- PEFT 0.13.2
- Transformers 4.47.0.dev0
- Pytorch 2.4.0+cu121
- Datasets 3.1.0
- Tokenizers 0.20.3