vidore/colqwen2.5-v0.2#

From Plugin

Note

This is a remotely-sourced model from the colqwen2_5_v0_2 plugin, maintained by the community. It is not part of FiftyOne core and may have special installation requirements. Please review the plugin documentation and license before use.

ColQwen is a model based on a novel model architecture and training strategy based on Vision Language Models (VLMs) to efficiently index documents from their visual features. It is a Qwen2.5-VL-3B extension that generates ColBERT- style multi-vector representations of text and images..

Details

  • Model name: vidore/colqwen2.5-v0.2

  • Model source: https://huggingface.co/vidore/colqwen2.5-v0.2

  • Model author: Vidore

  • Model license: Apache 2.0

  • Exposes embeddings? yes

  • Tags: classification, logits, embeddings, torch, visual-document-retrieval, zero-shot

Requirements

  • Packages: huggingface-hub, transformers, torch, torchvision, colpali-engine

  • CPU support

    • yes

  • GPU support

    • yes

Example usage

 1import fiftyone as fo
 2import fiftyone.zoo as foz
 3
 4foz.register_zoo_model_source("https://github.com/harpreetsahota204/colqwen2_5_v0_2")
 5
 6dataset = foz.load_zoo_dataset(
 7    "coco-2017",
 8    split="validation",
 9    dataset_name=fo.get_default_dataset_name(),
10    max_samples=50,
11    shuffle=True,
12)
13
14model = foz.load_zoo_model("vidore/colqwen2.5-v0.2")
15
16dataset.apply_model(model, label_field="predictions")
17
18session = fo.launch_app(dataset)