RichardErkhov/mikewang_-_PVD-160k-Mistral-7b-gguf

Quantization made by Richard Erkhov.

This repository contains a quantized version of the model presented in Visually Descriptive Language Model for Vector Graphics Reasoning.

Github

Discord

Request more models

Project page: https://mikewangwzhl.github.io/VDLM/ Code: https://github.com/MikeWangWZHL/VDLM

PVD-160k-Mistral-7b - GGUF

• Model creator: https://huggingface.co/mikewang/
• Original model: https://huggingface.co/mikewang/PVD-160k-Mistral-7b/

Name	Quant method	Size
PVD-160k-Mistral-7b.Q2_K.gguf	Q2_K	2.53GB
PVD-160k-Mistral-7b.IQ3_XS.gguf	IQ3_XS	2.81GB
PVD-160k-Mistral-7b.IQ3_S.gguf	IQ3_S	2.96GB
PVD-160k-Mistral-7b.Q3_K_S.gguf	Q3_K_S	2.95GB
PVD-160k-Mistral-7b.IQ3_M.gguf	IQ3_M	3.06GB
PVD-160k-Mistral-7b.Q3_K.gguf	Q3_K	3.28GB
PVD-160k-Mistral-7b.Q3_K_M.gguf	Q3_K_M	3.28GB
PVD-160k-Mistral-7b.Q3_K_L.gguf	Q3_K_L	3.56GB
PVD-160k-Mistral-7b.IQ4_XS.gguf	IQ4_XS	3.67GB
PVD-160k-Mistral-7b.Q4_0.gguf	Q4_0	3.83GB
PVD-160k-Mistral-7b.IQ4_NL.gguf	IQ4_NL	3.87GB
PVD-160k-Mistral-7b.Q4_K_S.gguf	Q4_K_S	3.86GB
PVD-160k-Mistral-7b.Q4_K.gguf	Q4_K	4.07GB
PVD-160k-Mistral-7b.Q4_K_M.gguf	Q4_K_M	4.07GB
PVD-160k-Mistral-7b.Q4_1.gguf	Q4_1	4.24GB
PVD-160k-Mistral-7b.Q5_0.gguf	Q5_0	4.65GB
PVD-160k-Mistral-7b.Q5_K_S.gguf	Q5_K_S	4.65GB
PVD-160k-Mistral-7b.Q5_K.gguf	Q5_K	4.78GB
PVD-160k-Mistral-7b.Q5_K_M.gguf	Q5_K_M	1.7GB
PVD-160k-Mistral-7b.Q5_1.gguf	Q5_1	5.07GB
PVD-160k-Mistral-7b.Q6_K.gguf	Q6_K	5.53GB
PVD-160k-Mistral-7b.Q8_0.gguf	Q8_0	7.17GB

Original model description:

license: apache-2.0 datasets: - mikewang/PVD-160K

Text-Based Reasoning About Vector Graphics

🌐 Homepage • 📃 Paper • 🤗 Data (PVD-160k) • 🤗 Model (PVD-160k-Mistral-7b) • 💻 Code

We observe that current large multimodal models (LMMs) still struggle with seemingly straightforward reasoning tasks that require precise perception of low-level visual details, such as identifying spatial relations or solving simple mazes. In particular, this failure mode persists in question-answering tasks about vector graphics—images composed purely of 2D objects and shapes.

To solve this challenge, we propose Visually Descriptive Language Model (VDLM), a visual reasoning framework that operates with intermediate text-based visual descriptions—SVG representations and learned Primal Visual Description, which can be directly integrated into existing LLMs and LMMs. We demonstrate that VDLM outperforms state-of-the-art large multimodal models, such as GPT-4V, across various multimodal reasoning tasks involving vector graphics. See our paper for more details.