---
base_model: Qwen/Qwen2.5-VL-7B-Instruct
license: apache-2.0
library_name: transformers
pipeline_tag: image-text-to-text
tags:
- qwen2.5-vl
- geolocation
- vision-language
- safetensors
---

# HoloGeo

HoloGeo is a Qwen2.5-VL-7B-Instruct based vision-language model for evidence-driven image geolocation.

This repository contains the merged BF16 model weights saved as `safetensors`. The LoRA adapter from the RL checkpoint has been merged into the base model, so the model can be loaded directly with `transformers`.

## Load

```python
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration

model_id = "PPKQ/HoloGeo"

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)
processor = AutoProcessor.from_pretrained(model_id)
```

## Notes

- Base model: `Qwen/Qwen2.5-VL-7B-Instruct`
- Checkpoint source: `RL_weights2/checkpoint-8000`
- Serialization: sharded `safetensors`
- Precision: BF16

The accompanying dataset is available at `https://huggingface.co/datasets/PPKQ/HoloGeo`.