AccuralAI Canonicalize

Canonicalization helpers and plugins for the AccuralAI LLM pipeline.

Overview

accuralai-canonicalize exposes canonicalization plugins that standardize input processing:

Standard Canonicalizer: Trims prompt whitespace, applies templates, normalizes tags
Advanced Canonicalizer: Additional processing with configurable metadata injection

Key Features

Prompt Normalization: Consistent whitespace and formatting
Template Application: Optional prompt templating
Tag Management: Normalization and deduplication of metadata tags
Cache Key Generation: Deterministic keys from configurable metadata fields

API Reference

AccuralAI Canonicalize

Usage Examples

Standard Canonicalizer

from accuralai_canonicalize.canonicalizer import build_standard_canonicalizer

canonicalizer = build_standard_canonicalizer(
    trim_whitespace=True,
    apply_templates=True,
    normalize_tags=True
)

result = await canonicalizer.canonicalize({
    "prompt": "  Hello, world!  ",
    "tags": ["test", "demo", "TEST"],
    "metadata": {"user": "developer"}
})

Advanced Canonicalizer

from accuralai_canonicalize.canonicalizer import build_advanced_canonicalizer

canonicalizer = build_advanced_canonicalizer(
    template_config={
        "system_template": "You are a helpful assistant: {prompt}",
        "user_template": "User: {prompt}"
    },
    metadata_defaults={
        "model": "default",
        "temperature": 0.7
    }
)

result = await canonicalizer.canonicalize({
    "prompt": "Hello",
    "role": "user"
})

Configuration

Canonicalizers are registered as entry points under accuralai_core.canonicalizers:

standard: Basic canonicalization with trimming and normalization
advanced: Extended canonicalization with templating and metadata injection

Development

# Install in development mode
pip install -e packages/accuralai-canonicalize[dev]

# Run tests
pytest packages/accuralai-canonicalize/tests/