---
language: en
license: apache-2.0
library_name: transformers
tags:
- qwen3.5
- unsloth
- autonomous-driving
- driving-scene
- cocreator
- vision-language
datasets:
- NIyueeE/cocreator-driving-scene
pipeline_tag: image-text-to-text
---

# Qwen3.5-0.8B-cocreator

**Qwen3.5-0.8B** fine-tuned on the [CoCreator Driving Scene](https://huggingface.co/datasets/NIyueeE/cocreator-driving-scene) dataset for driving scene causal understanding.

## Model Details

- **Base model:** [Qwen/Qwen3.5-0.8B](https://huggingface.co/Qwen/Qwen3.5-0.8B)
- **Dataset:** [NIyueeE/cocreator-driving-scene](https://huggingface.co/datasets/NIyueeE/cocreator-driving-scene) - 1,227 driving scene samples, each with multi-frame video and causal text descriptions
- **Fine-tuning method:** QLoRA (4-bit) via Unsloth
- **Vision:** Native multimodal (image+text)

## Training

### Platform

Google Colab ([colab.research.google.com](https://colab.research.google.com/)) with NVIDIA A100-SXM4-40GB.

### Training Log

```
Unsloth 2026.5.5: Fast Qwen3_5 patching. Transformers: 5.5.0.
NVIDIA A100-SXM4-40GB. Num GPUs = 1. Max memory: 39.494 GB.
Torch: 2.10.0+cu128. CUDA: 8.0. CUDA Toolkit: 12.8. Triton: 3.6.0
Bfloat16 = TRUE. FA [Xformers = 0.0.35. FA2 = False]

Num examples = 1,227 | Num Epochs = 7 | Total steps = 50
Batch size per device = 128 | Gradient accumulation steps = 1
Total batch size (128 x 1 x 1) = 128
Trainable parameters = 13,181,952 of 866,167,872 (1.52% trained)
```

### Loss Curve

![Training loss](training_dashboard.png)

### Training Script

See [`finetune_cocreator_coclab.ipynb`](./finetune_cocreator_coclab.ipynb) for the complete fine-tuning notebook.

### Hyperparameters

| Parameter | Value |
|-----------|-------|
| LoRA r | 16 |
| LoRA alpha | 16 |
| LoRA dropout | 0 |
| Target modules | all-linear |
| Fine-tuned layers | vision + language + attention + MLP |
| Optimizer | adamw_8bit |
| Learning rate | 5e-5 (cosine schedule) |
| Max steps | 50 |
| Epochs | 7 |
| Gradient checkpointing | unsloth |
| Resolution | 800×450 (resized) |

## Usage

```python
from transformers import AutoModel, AutoTokenizer
import torch

model = AutoModel.from_pretrained(
    "NIyueeE/Qwen3.5-0.8B-cocreator",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
    "NIyueeE/Qwen3.5-0.8B-cocreator",
    trust_remote_code=True,
)
```

## Intended Use

This model is fine-tuned for driving scene causal understanding. It takes multi-frame driving images as input and generates causal relationship text descriptions. The primary use case is as a feature extractor in the ReCogDrive autonomous driving VLA training pipeline.

## License

Apache 2.0