Open Source Today (2024-07-30): SciPhi AI LLM, 98% Cheaper than GPT-4
Explore cutting-edge AI projects like Triplex for knowledge graphs, SAM 2 for segmentation, TaskGen for task automation, and more. Enhance your AI toolkit today!
Here are some interesting AI open-source models and frameworks I wanted to share today:
Project: Triplex
Triplex, developed by SciPhi.AI, is an advanced open-source LLM for creating knowledge graphs from unstructured data.
It works by extracting triples (simple statements consisting of a subject, predicate, and object) from text or other data sources.
Compared to GPT-4, Triplex reduces the cost of creating knowledge graphs by 98% and can be deployed locally to ensure data privacy.
https://huggingface.co/SciPhi/Triplex
Project: SAM 2
SAM 2 is the next-generation Segment Anything Model (SAM) from Meta, capable of object segmentation in videos and images.
By treating images as single-frame videos, SAM 2 extends video processing capabilities. The model uses a simple Transformer architecture and streaming memory for real-time video processing.
The SA-V dataset, the largest visual segmentation dataset to date, was built through user interactions. SAM 2 performs excellently across various tasks and visual domains.
https://github.com/facebookresearch/segment-anything-2
Project: TaskGen
TaskGen is a task-based agentic framework using StrictJSON, designed to provide efficient solutions by breaking tasks into subtasks.
The framework supports single-agent and meta-agent modes, integrates LLM and external functions, and offers multi-modality support, memory functions, and global context configuration.
TaskGen uses JSON format for chain-of-thought prompting, ensuring the correctness and format of output fields.
https://github.com/simbianai/taskgen
Project: RapidLayout
RapidLayout is a project for analyzing Chinese and English document layouts. It can analyze given document images (like screenshots of papers and reports) and locate titles, paragraphs, tables, and images.
The project supports layout analysis in various scenarios, including tables, English documents, Chinese documents, papers, and reports.
Users can quickly invoke the project's functions through simple Python scripts or command-line tools.
https://github.com/RapidAI/RapidLayout
Project: Knowledge Graph Maker
Knowledge Graph Maker is a Python tool that can turn any text into a knowledge graph using a given ontology.
The library consistently creates graphs and is resilient to incorrect responses generated by large language models.
https://github.com/rahulnyk/knowledge_graph_maker
Project: Awesome-Production-LLM
Awesome-Production-LLM is a curated list of open-source libraries focused on large language models (LLMs) in production environments.
The project includes open-source projects for data preprocessing, model training, evaluation, inference, application development, and security.