Skip to content

Local llms for data visualisation c

You asked if there are any large language models (LLMs) that could run on your hardware and be useful for data visualization, specifically for generating graphs from natural language.

Hardware Overview

Based on the specifications of your workstation:

  • CPU: Intel Core i7-12700F (12 cores, 25MB cache)
  • GPU: AMD Radeon RX 7700 XT (12GB VRAM)
  • RAM: 64GB DDR5
  • Storage: NVMe SSDs (1.1TB + additional SSD storage)
  • OS: OpenSUSE Tumbleweed with KDE Plasma

This is a powerful setup, particularly with the 12GB VRAM GPU and 64GB of RAM, which are critical for running LLMs locally.

LLMs for Data Visualization

Several LLMs can assist with generating graphs and data visualizations from natural language queries. Here are some options that could work well on your hardware:

1. GPT-based Models (e.g., GPT-4, GPT-NeoX)

  • Use Case: These models can be prompted to generate code for data visualizations (e.g., using Python libraries like Matplotlib or Plotly) based on natural language instructions.
  • Hardware Fit: Your GPU's 12GB VRAM is suitable for running inference on smaller versions of GPT models locally. For larger models (like GPT-3 or GPT-4), you may need to rely on cloud-based APIs unless you use optimized versions like GPT-J or GPT-NeoX.
  • Data Visualization Capability: These models can generate code to create various types of charts (bar charts, line charts, scatter plots) by interpreting natural language prompts[6].

2. LLaMA Models (e.g., LLaMA 2)

  • Use Case: LLaMA models are efficient and can be fine-tuned or used for generating code that creates visualizations.
  • Hardware Fit: With optimizations like quantization, LLaMA models can run efficiently even on consumer-grade GPUs like yours.
  • Data Visualization Capability: While not specialized in graph generation, they can produce Python code for visualization libraries based on your input[6].

3. InstructGLM

  • Use Case: InstructGLM is a model designed to handle graph-related tasks by understanding both the structure and semantics of graphs described in natural language.
  • Hardware Fit: This model is relatively lightweight compared to other LLMs, making it feasible to run locally with your hardware[7].
  • Data Visualization Capability: InstructGLM can help in generating knowledge graphs and performing node classification or link prediction tasks from textual descriptions[7].

4. Hugging Face Transformers (e.g., BERT, T5)

  • Use Case: Hugging Face offers several pre-trained models that can be fine-tuned for specific tasks like data visualization.
  • Hardware Fit: These models can run on your system using frameworks like PyTorch or TensorFlow. Your system's GPU and RAM should handle these models well.
  • Data Visualization Capability: You can use these models to generate code that produces visualizations based on natural language queries[2][5].

Tools for Data Visualization

Several tools integrate LLMs to assist with graph creation from natural language:

  1. Neo4j + LLMs: NebulaGraph’s Text2Cypher technology allows users to query graph databases using natural language, which could be useful for creating knowledge graphs from unstructured text[8].
  2. Ontotext’s GraphDB: This tool integrates LLMs to allow querying knowledge graphs via natural language interfaces, making it easier to visualize relationships between entities[9].

Conclusion

Your hardware is more than capable of running smaller or optimized versions of LLMs locally that can assist with data visualization tasks. Models like GPT-NeoX, InstructGLM, or Hugging Face transformers are good candidates for generating code that produces visualizations from natural language prompts. Additionally, tools like Neo4j or Ontotext’s GraphDB provide specialized solutions for building and querying knowledge graphs using LLMs.

The above text was generated by a large language model (LLM) and its accuracy has not been validated. This page is part of 'LLMs-on-LLMs,' a Github repository by Daniel Rosehill which explores how curious humans can use LLMs to better their understanding of LLMs and AI. However, the information should not be regarded as authoritative and given the fast pace of evolution in LLM technology will eventually become deprecated. This footer was added at 16-Nov-2024.