artifex 0.4.0

Create your private AI model with no training data or GPUs 🤖🚀.

Artifex

Documentation | Tutorial

🎯 Create Task-Specific LLMs • 📊 No training data needed • 🌱 No GPU needed • 🖥️ CPU Inference & Fine-Tuning

---

Artifex is a Python library for:

1. Using small, pre-trained task-specific LLMs locally on CPU
2. Fine-tuning them on CPU without any training data — just based on your instructions for the task at hand.

At this time, we support 7 main tasks:

• 🛡️ Guardrail: Flags unsafe, harmful, or off-topic messages.
• 🗣️ Intent Classification: Classifies user messages into predefined intent categories.
• 🔀 Reranker: Ranks a list of items or search results based on relevance to a query.
• 🙂 Sentiment Analysis: Determines the sentiment (positive, negative, neutral) of a given text.
• 😡 Emotion Detection: Identifies the emotion expressed in a given text.
• 🏷️ Named Entity Recognition (NER): Detects and classifies named entities in text (e.g., persons, organizations, locations).
• 🥸 Text Anonymization: Removes personally identifiable information (PII) from text.

For each task, Artifex provides three easy-to-use APIs:

1. Inference API to use a default, pre-trained small LLM to perform that task out-of-the-box locally on CPU.
2. Fine-tune API to fine-tune the default model based on your requirements, without any training data and on CPU. The fine-tuned model is generated on your machine and is yours to keep.
3. Load API to load your fine-tuned model locally on CPU, and use it for inference or further fine-tuning.

We will be adding more tasks soon, based on user feedback. Want Artifex to perform a specific task? Suggest one or vote one up.

🔥 How does it work?

Problem

LLMs available on the market can be broadly classified into two categories:

• General-purpose LLMs (GPT, Claude, Llama, etc.) have two main limitations:

  1. They are designed for open-ended tasks, which makes them overkill and often suboptimal for simpler, specific use cases.
  2. If open-source, they require expensive GPUs for training and inference; if not open-source, they incur high costs for usage via APIs and have data privacy concerns since your data is sent to 3rd-party servers.

• Smaller LLMs (DistilBERT, TinyBERT, etc.) can sometimes be trained and run locally on CPU, but they require large amounts of labeled training data to perform well on specific tasks — which is often not available.

Solution

Artifex overcomes these limitations by enabling you to:

• Use small (capped at 500 Mb in size), pre-trained task-specific LLMs locally on CPU, thereby eliminating costs and data privacy concerns.

• Fine-tune these models based on your requirements, without any training data — just based on your instructions for the task at hand — thereby obtaining higher accuracy on your specific use case.

  How is it possible?

  Artifex generates synthetic training data on-the-fly based on your instructions, and uses this data to fine-tune small LLMs for your specific task. This approach allows you to create effective models without the need for large labeled datasets.

🚀 Quick Start

Install Artifex with:

pip install artifex

🛡️ Guardrail Model

Use the default Guardrail model (inference API)

Need a general-purpose guardrail model? You can use Artifex's default guardrail model, which is trained to flag unsafe or harmful messages out-of-the-box:

from artifex import Artifex

guardrail = Artifex().guardrail
print(guardrail("How do I make a bomb?"))

# >>> [{'label': 'unsafe', 'score': 0.9976}]

Learn more about the default guardrail model and what it considers safe vs unsafe on our Guarderail HF model page.

Create & use a custom Guardrail model (fine-tune & load APIs)

Need more control over what is considered safe vs unsafe? Fine-tune your own guardrail model, use it locally on CPU and keep it forever:

from artifex import Artifex

guardrail = Artifex().guardrail

model_output_path = "./output_model/"

guardrail.train(
    instructions=[
        "Discussing a competitor's products or services is not allowed.",
        "Sharing our employees' personal information is prohibited.",
        "Providing instructions for illegal activities is forbidden.",
        "Everything else is allowed.",
    ],
    output_path=model_output_path
)

guardrail.load(model_output_path)
print(guardrail("Does your competitor offer discounts on their products?"))

# >>> [{'label': 'unsafe', 'score': 0.9970}]

🗣️ Intent Classification model

Use the default Intent Classification model (inference API)

Need a general-purpose intent classification model? You can use Artifex's default intent classification model, which is trained to recognize common intents out-of-the-box:

from artifex import Artifex

intent_classifier = Artifex().intent_classifier

print(intent_classifier("Hey there, how are you doing?"))

# >>> [{'label': 'greeting', 'score': 0.9955}]

Learn more about the default intent classification model and what intents it is trained to recognize on our Intent Classification HF model page.

Create & use a custom Intent Classification model (fine-tune & load APIs)

Need more control over the classes recognized, or do you want to tailor the model to your specific domain for better results? Fine-tune your own intent classification model, use it locally on CPU and keep it forever:

from artifex import Artifex

intent_classifier = Artifex().intent_classifier

model_output_path = "./output_model/"

intent_classifier.train(
    domain="e-commerce customer support",
    classes={
        "order_status": "Inquiries about the status of an order.",
        "return_item": "Requests to return a purchased item.",
        "product_info": "Questions about product details or specifications.",
        "greeting": "Friendly greetings or salutations.",
    },
    output_path=model_output_path
)

intent_classifier.load(model_output_path)
print(intent_classifier("I want to return an item I bought last week."))

# >>> [{'label': 'return_item', 'score': 0.9914}]

🔀 Reranker model

Use the default Reranker model (inference API)

Need a general-purpose reranker model? You can use Artifex's default reranker model, which is trained to rank items based on relevance out-of-the-box:

from artifex import Artifex

reranker = Artifex().reranker

print(reranker(
    query="Best programming language for data science",
    documents=[
        "Java is a versatile language typically used for building large-scale applications.",
        "Python is widely used for data science due to its simplicity and extensive libraries.",
        "JavaScript is primarily used for web development.",
    ]
))

# >>> [('Python is widely used for data science due to its simplicity and extensive libraries.', 3.8346), ('Java is a versatile language typically used for building large-scale applications.', -0.8301), ('JavaScript is primarily used for web development.', -1.3784)]

Create & use a custom Reranker model (fine-tune & load APIs)

Want to fine-tune the Reranker model on a specific domain for better accuracy? Fine-tune your own reranker model, use it locally on CPU and keep it forever:

from artifex import Artifex

reranker = Artifex().reranker

model_output_path = "./output_model/"

reranker.train(
    domain="e-commerce product search",
    output_path=model_output_path
)

reranker.load(model_output_path)
print(reranker(
    query="Laptop with long battery life",
    documents=[
        "A powerful gaming laptop with high-end graphics and performance.",
        "An affordable laptop suitable for basic tasks and web browsing.",
        "This laptop features a battery life of up to 12 hours, perfect for all-day use.",
    ]
))

# >>> [('This laptop features a battery life of up to 12 hours, perfect for all-day use.', 4.7381), ('A powerful gaming laptop with high-end graphics and performance.', -1.8824), ('An affordable laptop suitable for basic tasks and web browsing.', -2.7585)]

🔣 Other Tasks

For more details and examples on how to use Artifex for the other available tasks, check out the Available Tasks section below and our Documentation.

🔧 Available Tasks & Examples

Task	Default Model	Default & Fine-Tuned Model Size	CPU Inference	CPU Fine-Tuning	Code Examples
🛡️ Guardrail	tanaos/tanaos-guardrail-v1	0.1B params, 500Mb	✅	✅	Examples
🗣️ Intent Classification	tanaos/tanaos-intent-classifier-v1	0.1B params, 500Mb	✅	✅	Examples
🔀 Reranker	cross-encoder/mmarco-mMiniLMv2-L12-H384-v1	0.1B params, 470Mb	✅	✅	Examples
🙂 Sentiment Analysis	tanaos/tanaos-sentiment-analysis-v1	0.1B params, 470Mb	✅	✅	Examples
😡 Emotion Detection	tanaos/tanaos-emotion-detection-v1	0.1B params, 470Mb	✅	✅	Examples
🏷️ Named Entity Recognition	tanaos/tanaos-NER-v1	0.1B params, 500Mb	✅	✅	Examples
🥸 Text Anonymization	tanaos/tanaos-text-anonymizer-v1	0.1B params, 500Mb	✅	✅	Examples

🤝 Contributing

Contributions are welcome! Whether it's a new task module, improvement, or bug fix — we’d love your help. Not ready to contribute code? You can also help by suggesting a new task or voting up any suggestion.

git clone https://github.com/tanaos/artifex.git
cd artifex
pip install -e .

Before making a contribution, please review the CONTRIBUTING.md and CLA.md, which include important guidelines for contributing to the project.

📚 Documentation & Support

• Full documentation: https://docs.tanaos.com/artifex
• Get in touch: [e-mail]