RainbirdAI/Noesis-1.0-8bit
Model Overview
Noesis is an open source, fine-tuned large language model with accompanying APIs and examples, designed to convert unstructured documentation into structured, executable knowledge graphs compatible with the Rainbird symbolic reasoning engine.
By automating the creation of ontologies and inference rules from textual data, Noesis creates deterministic, explainable AI guardrails that seamlessly integrate into existing machine learning pipelines.
This initial release is optimized for Apple's MLX framework, with support for additional platforms coming very soon.
About Rainbird
Rainbird is the enterprise platform that backs Noesis. It includes our powerful symbolic inference engine that can reason over your graphs, and management tools that enable you to visualise, amend, test, version and report against them. While you do not need to use the Rainbird management platform, it will enable you to see and work more easily with the graphs that Noesis has produced.
The Rainbird Community platform is free for non-commercial purposes. It provides you the ability to visually inspect your Noesis-generated graphs as well as test and amend, version, deploy and report against your graphs. It also includes access to valuable training resources.
You can sign up for free access to Rainbird Comunity here.
Our solution engineers can provide dedicated support for integrating Noesis into high-stakes applications, ensuring your AI systems meet regulatory standards while maintaining the development velocity modern organizations demand.
If you have specific questions, please contact us.
Why Noesis & Rainbird
We solve the fundamental challenge of LLM hallucinations and inconsistency by transforming organizational documentation and expertise into sophisticated knowledge graphs that deliver precise, explainable outcomes every-time when reasoned over by our symbolic reasoning engine. We combine the linguistic flexibility of LLMs with graph-based deterministic reasoning that ensures every decision follows explicit regulation, policy or procedure.
The two patterns of Rainbird deployment are: 1) Graph-First Reasoning, where the deterministic engine makes decisions and the LLM handles communication, and 2) Post-Generation Validation, where the LLM generates responses that are verified by the deterministic engine before delivery.
A white paper that describes typical deployment patterns is available here.
As a developer, you'll benefit from being able to automate knowledge graph creation and integrate deterministic reasoning into your AI applications in minutes, not weeks.
- Automated Knowledge Capture: Convert regulations, policies, and tacit expertise into deterministic models without manual knowledge engineering or coding
- Deterministic Reasoning: Eliminate probabilistic uncertainty by enabling LLM-generated outputs to be passed through a sophisticated deterministic guardrail, or alternatively use it as the primary reasoner.
- Causal Explainability: Provide complete audit trails that detail how each decision was reached
- Integration Flexibility: Seamlessly incorporate into existing workflows and systems
- Developer-First Approach: Open-source foundation with a free tier of the Rainbird platform backed by enterprise upgrade options
Python API & Pipeline Integration
To facilitate integration into Python-based workflows, Noesis is accompanied by a modular pipeline framework available at NoesisAPI. This framework allows developers to:
- Pre and Post Processing: Use LLMs to prepare a large body of text for Noesis processing, and to validate the graph produced.
- Command line tooling: Comand line tools to take PDF documentation and create
- Integrate with Rainbird: Send data to the Rainbird API for knowledge graph generation.
- Extend Functionality: Create custom processing steps to handle unique requirements.
Installation
# Clone the repository
git clone https://github.com/benjamta/NoesisAPI.git
cd NoesisAPI
# Install dependencies
pip install -r requirements.txt
# Install the package in development mode
pip install -e .
Quick Start Example
from rainbird import Noesis
# Create a pipeline with custom configuration
api = Noesis(config={
"noesis_model": "path/to/model",
"temperature": 0.7,
"max_tokens": 5000
})
# Configure and process text
api.configure()
result = api.process("People speak the language of the country they're born in")
print(result)
This example demonstrates how to process a simple sentence into a knowledge graph using the default pipeline configuration.
Command Line tooling
We also provide a command line tool for procesisng PDF files through the Noesis pipeline, resulting in a new graph in your Rainbrid account ready for you to use.
This application can be configured to use a local LLM for pre and post processing, or can use the Anthropic API. You will need your Rainbrd API key (available from you Rainbird Studio account) to use this tool.
Installation
- Clone the repository and install requirements:
# Clone the repository
git clone https://github.com/benjamta/NoesisAPI.git
cd NoesisAPI
# Install dependencies
pip install -r requirements.txt
# Install the package in development mode
pip install -e .
- Create a
.envfile in the project root directory and add your Rainbird API key and optionally your Anthropic API key. THe Anthropic key can be omitted if you're using a lcoak LLM for pre and post processing:
RAINBIRD_API_KEY=you_rainbird_api_key_here
ANTHROPIC_API_KEY=your_api_key_here
- The
config.yamlfile in the project root directory includes sample configuration settings. You should configure this for your setup.
Usage
Basic usage:
python noesis_process.py path/to/your/file.pdf
Command Line Options
--config: Specify a custom config file (default: config.yaml)--anthropic-api-key: Override the Anthropic API key--rainbird-anthropic-api-key: Specify a different API key for Rainbird error correction--graph-name: Override the graph name template
For each step (noesis, preprocess, validate, rainbird), you can override:
--{step}-model: Model path/name--{step}-type: Model type (local or anthropic)--{step}-anthropic-model: Anthropic model name--{step}-adapter-path: Adapter path--{step}-temperature: Temperature setting--{step}-max-tokens: Maximum tokens
Platform Support
- Current: Apple MLX
- Upcoming: Support for additional platforms is under development and will be released in future updates.
Resources
- Downloads last month
- 9
Model tree for RainbirdAI/Noesis-1.0-bf16
Base model
meta-llama/Llama-3.1-70B