---
license: apache-2.0
datasets:
- leon-se/ForestFireInsights-Eval
language:
- en
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
tags:
- climate
---

# Model Card: ForestFireVLM-7B-FP8-Dynamic

## Model Description
ForestFireVLM-7B is a specialized vision-language model fine-tuned from [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) specifically for forest fire detection and analysis tasks. This model is designed to identify and analyze various aspects of forest fires from aerial imagery, including smoke detection, flame visibility, fire characteristics, and potential hazards.
The [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory/) framework was used for fine-tuning this model. Quantized to FP8 with [llm-compressor](https://github.com/vllm-project/llm-compressor).

## How to Use
The model can be used with vLLM to create an OpenAI API compatible endpoint:

```bash
vllm serve leon-se/ForestFireVLM-7B-FP8-Dynamic --max-model-len 10000
```

## Evaluations
Evaluations were done with our code available on [GitHub](https://github.com/leon-seidel/ForestFireVLM), using the [ForestFireInsights-Eval](https://huggingface.co/datasets/leon-se/ForestFireInsights-Eval) dataset.

## Citations

This model is associated with research currently under peer review with MDPI. Please cite our paper when using this model:

```
[Citation will be added when the paper is published]
```