---
library_name: peft
license: other
base_model: Qwen/Qwen3-VL-8B-Instruct
tags:
- base_model:adapter:Qwen/Qwen3-VL-8B-Instruct
- llama-factory
- lora
- transformers
pipeline_tag: text-generation
model-index:
- name: aitw_reasoning_tf_raw_rl
  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. -->

# aitw_reasoning_tf_raw_rl

This model is a fine-tuned version of [Qwen/Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) on the aitw_general_reasoning_tf_raw_rl_think_coord_only and the aitw_single_reasoning_tf_raw_rl_think_coord_only datasets.
It achieves the following results on the evaluation set:
- Loss: 0.5224

## 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: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- total_eval_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 0.05
- num_epochs: 1.0

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 0.5424        | 0.2638 | 500  | 0.5382          |
| 0.5389        | 0.5276 | 1000 | 0.5276          |
| 0.5166        | 0.7915 | 1500 | 0.5238          |


### Framework versions

- PEFT 0.18.1
- Transformers 5.2.0
- Pytorch 2.10.0+cu128
- Datasets 4.0.0
- Tokenizers 0.22.2