← Catalog

vidore/colqwen2.5-v0.2

Add to compare Open comparison →

Primitive: /encode · Encode · Qwen2

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.

View on Hugging Face → Fine-tuned from vidore/colqwen2.5-base

Overview

Hardware: L4RTX-PRO-6000 — drives latency, throughput & cost

Size	7.0B params
Tasks	/encode
License	mit
Languages	en
Latency	1.9 s
Throughput	7.6 mpix/s
Cost	— /1M tok

Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.

Embedding

Output types	Multi-Vec
Dimensions	multivector: 128
Max sequence length	2,048
Inputs	text · image

Benchmarks