Artifex is a versatile Python library designed for crafting task-specific Large Language Models (LLMs) without the need for prior training data, making it an ideal solution for offline Natural Language Processing (NLP) and text classification tasks. This innovative tool can create small, efficient models that can be deployed locally, which not only streamlines development but also enhances privacy and performance.

Key Features

• Create Tailored LLMs: Unlike general-purpose models that tackle a wide array of tasks, Artifex enables the development of models focused on specific applications, enhancing accuracy and relevance.
• No Training Data Required: The library utilizes synthetic data generation methods, eliminating the significant resource needs typically associated with training data preparation.
• Resource Efficient: Models built with Artifex operate seamlessly on CPUs, negating the need for GPU resources which leads to cost savings and broader accessibility.
• API Cost Reduction: By offloading tasks to locally run models, Artifex can significantly reduce reliance on costly general-purpose LLM API calls.
• Prebuilt Templates: The library comes equipped with templates for common use cases, including Guardrail and Intent Classification models, with additional options to be released based on user community requests.

Offline NLP Applications

Models developed using Artifex support various offline NLP tasks such as:

• Guardrail Model: Monitors and flags unsafe or irrelevant messages, ensuring interactions remain appropriate.
• Intent Classification: Efficiently categorizes user messages to support task automation or routing.

Getting Started: Usage Examples

Create a Guardrail Model To establish a guardrail model, which helps in managing communication standards, follow these steps:

from artifex import Artifex

guardrail = Artifex().guardrail

guardrail.train(
    instructions=[
        "Soft medical advice is allowed, but should be general.",
        "Discussions about cosmetic products are permitted.",
        "Hard medical advice is prohibited in all cases."
    ]
)

Implementing the Model Replace API calls with your local model for message safety checks:

# Load your locally trained guardrail model
.guardrail.load("artifex_output/run-<timestamp>/output_model/")
is_safe = guardrail(user_message)

Create an Intent Classification Model To train a model for classifying intents based on user input:

from artifex import Artifex

intent_classifier = Artifex().intent_classifier

intent_classifier.train(
    classes={
        "send_email": "Intent to send an email.",
        "schedule_meeting": "Intent to set up a meeting.",
        "cancel_meeting": "Intent to cancel a meeting.",
        "reschedule_meeting": "Intent to change a meeting's time.",
    }
)

label = intent_classifier("I forgot to set up that meeting with John.")
print(label)  # Outputs the classified intent

Explore More

Discover more functionalities, tutorials, and use cases through the comprehensive documentation and community resources.

Artifex is positioned to significantly streamline and enhance the proficiency of NLP tasks by enabling developers to create specialized models efficiently, ultimately paving the way for more effective and relevant AI applications.