ColFlor: Towards BERT-Size Vision-Language Document Retrieval Models
•
16
Ahmed Masry PRO
ahmed-masry
AI & ML interests
Multimodal Chart Understanding,
Multimodal Document AI,
Multimodal Vision - Language Models,
Recent Activity
authored
a paper
15 days ago
BigDocs: An Open and Permissively-Licensed Dataset for Training
Multimodal Models on Document and Code Tasks
new activity
about 2 months ago
ahmed-masry/ColFlor:Update README.md
Articles
Organizations
None yet
Posts
3
Post
1270
🚀 Introducing ColFlor: An Efficient, OCR-Free Vision-Language Document Retrieval Model 🌟
Earlier this year, ColPali revolutionized document retrieval by eliminating the need for error-prone OCR pipelines. Instead, it directly processes the document images. However, with its 3 billion parameters, ColPali is computationally heavy for large-scale applications.
That’s where ColFlor comes in—a smaller, faster alternative! 🎉 At 17x smaller than ColPali, ColFlor offers a more efficient, OCR-free document retrieval solution, making it ideal for users with limited computing resources (GPU Poor). 💡
Key Highlights:
🧠 174M parameters (vs. 3B for ColPali)
⚡ 9.8x faster query encoding, 5.25x faster image encoding
📉 Only 1.8% performance drop on text-rich English documents
Check out the full blog post for more insights on modeling, training, and evaluations across various document retrieval tasks! 🚀
Also, feel free to try our demo on huggingface 🤗
🔗 Resources:
📄 Blog post: https://huggingface.co./blog/ahmed-masry/colflor
🧠 Model: ahmed-masry/ColFlor
💻 Demo: ahmed-masry/ColFlor-Demo
🏋️♂️ Training code: https://github.com/AhmedMasryKU/colflor
📊 Evaluation code: https://github.com/AhmedMasryKU/vidore-benchmark-colflor
Earlier this year, ColPali revolutionized document retrieval by eliminating the need for error-prone OCR pipelines. Instead, it directly processes the document images. However, with its 3 billion parameters, ColPali is computationally heavy for large-scale applications.
That’s where ColFlor comes in—a smaller, faster alternative! 🎉 At 17x smaller than ColPali, ColFlor offers a more efficient, OCR-free document retrieval solution, making it ideal for users with limited computing resources (GPU Poor). 💡
Key Highlights:
🧠 174M parameters (vs. 3B for ColPali)
⚡ 9.8x faster query encoding, 5.25x faster image encoding
📉 Only 1.8% performance drop on text-rich English documents
Check out the full blog post for more insights on modeling, training, and evaluations across various document retrieval tasks! 🚀
Also, feel free to try our demo on huggingface 🤗
🔗 Resources:
📄 Blog post: https://huggingface.co./blog/ahmed-masry/colflor
🧠 Model: ahmed-masry/ColFlor
💻 Demo: ahmed-masry/ColFlor-Demo
🏋️♂️ Training code: https://github.com/AhmedMasryKU/colflor
📊 Evaluation code: https://github.com/AhmedMasryKU/vidore-benchmark-colflor
Post
3476
📢 Exciting News! Our latest paper "ChartGemma" is out! 📊
🧵1/3: ChartGemma overcomes existing chart models key limitations that rely too much on data tables. Instead, it is trained on data generated directly from chart images, capturing crucial visual trends📸🔍
🧵2/3: ChartGemma builds upon PaliGemma from Google Research and is fine-tuned on a high-quality visual instruction tuning dataset generated from Gemini Flash 1.5. 🌟📊
🧵3/3: Achieves state-of-the-art results in chart summarization, question answering, and fact-checking tasks. 🏅📊 It can also generate more accurate and realistic chart summaries. 📝🔍
Our model and data are publicly available. We also have a cool web demo. Check it out! 🚀
Demo: ahmed-masry/ChartGemma
Code: https://github.com/vis-nlp/ChartGemma
Paper: ChartGemma: Visual Instruction-tuning for Chart Reasoning in the Wild (2407.04172)
🧵1/3: ChartGemma overcomes existing chart models key limitations that rely too much on data tables. Instead, it is trained on data generated directly from chart images, capturing crucial visual trends📸🔍
🧵2/3: ChartGemma builds upon PaliGemma from Google Research and is fine-tuned on a high-quality visual instruction tuning dataset generated from Gemini Flash 1.5. 🌟📊
🧵3/3: Achieves state-of-the-art results in chart summarization, question answering, and fact-checking tasks. 🏅📊 It can also generate more accurate and realistic chart summaries. 📝🔍
Our model and data are publicly available. We also have a cool web demo. Check it out! 🚀
Demo: ahmed-masry/ChartGemma
Code: https://github.com/vis-nlp/ChartGemma
Paper: ChartGemma: Visual Instruction-tuning for Chart Reasoning in the Wild (2407.04172)
spaces
6
Running
on
Zero
4
🏆
ColFlor Demo
Sleeping
📊
ChartInstruct FlanT5 XL
Running
on
Zero
5
🐢
ChartInstruct LLama2
ChartInstruct for general purpose chart understanding tasks
Running
on
Zero
1
🏆
UniChart Base
UniChart Base (Best for Chart Captioning & Chart to Table)
Running
on
Zero
101
🐨
ChartGemma
Running
🏃
UniChart ChartQA
UniChart finetuned on the ChartQA dataset
models
10
ahmed-masry/ColFlor
Updated
•
727
•
18
ahmed-masry/chartgemma
Image-Text-to-Text
•
Updated
•
2.35k
•
35
ahmed-masry/ChartInstruct-FlanT5-XL
Text2Text Generation
•
Updated
•
91
ahmed-masry/ChartInstruct-LLama2
Image-Text-to-Text
•
Updated
•
1.1k
•
2
ahmed-masry/unichart-base-960
Image-Text-to-Text
•
Updated
•
216
•
3
ahmed-masry/Chart-Mask-RCNN
Updated
ahmed-masry/unichart-opencqa-960
Image-Text-to-Text
•
Updated
•
810
ahmed-masry/unichart-chart2text-pew-960
Image-Text-to-Text
•
Updated
•
149
ahmed-masry/unichart-chart2text-statista-960
Image-Text-to-Text
•
Updated
•
63
ahmed-masry/unichart-chartqa-960
Image-Text-to-Text
•
Updated
•
355
•
2
datasets
7
ahmed-masry/unichart-qa-data
Viewer
•
Updated
•
300k
•
121
•
2
ahmed-masry/unichart-table-data
Viewer
•
Updated
•
305k
•
129
•
3
ahmed-masry/ChartGemma
Viewer
•
Updated
•
163k
•
997
•
8
ahmed-masry/ChartQA
Viewer
•
Updated
•
32.7k
•
338
•
20
ahmed-masry/chartqa_without_images
Viewer
•
Updated
•
32.7k
•
100
ahmed-masry/unichart-pretrain-data
Viewer
•
Updated
•
6.9M
•
86
•
4
ahmed-masry/UniChart-pretrain-images
Preview
•
Updated
•
69
•
1