alquimia-fair-forge 3.0.0

Alquimia Fair Forge library - Performance measurement for AI models and assistants

Performance-measurement library for evaluating AI models and assistants

Documentation • GitHub

What is Fair Forge?

Fair Forge provides comprehensive metrics for evaluating AI systems:

• Toxicity - Detect toxic language patterns with DIDT fairness framework
• Bias - Analyze biases across protected attributes (gender, race, religion, etc.)
• Context - Assess how well responses align with provided context
• Conversational - Evaluate dialogue quality using Grice's maxims
• Humanity - Measure how natural responses are through emotional analysis
• BestOf - Tournament-style comparison to find the best response
• Agentic - Evaluate autonomous agent behavior and tool usage

Quick Start

# Install with pip
pip install alquimia-fair-forge

# Or install specific modules
pip install "alquimia-fair-forge[toxicity]"
pip install "alquimia-fair-forge[bias]"
pip install "alquimia-fair-forge[all]"

from fair_forge import Retriever, Dataset
from fair_forge.metrics.toxicity import Toxicity

class MyRetriever(Retriever):
    def load_dataset(self) -> list[Dataset]:
        # Load your dataset here
        pass

metrics = Toxicity.run(MyRetriever, verbose=True)

Documentation

For complete documentation, guides, and API reference visit:

https://fairforge.alquimia.ai

Contributing

Setup

# Clone and install dependencies
git clone https://github.com/Alquimia-ai/fair-forge.git
cd fair-forge
uv sync

# Install pre-commit hooks (required)
uv run pre-commit install --hook-type pre-commit --hook-type commit-msg

Pre-commit Hooks

This project uses pre-commit to enforce code quality and commit conventions. The following hooks run automatically:

Hook	Stage	Purpose
trailing-whitespace	pre-commit	Remove trailing whitespace
end-of-file-fixer	pre-commit	Ensure files end with a newline
check-yaml / check-toml	pre-commit	Validate config files
check-added-large-files	pre-commit	Block files >1MB
detect-private-key	pre-commit	Prevent accidental secret commits
ruff	pre-commit	Lint and auto-fix Python code
ruff-format	pre-commit	Format Python code
commitizen	commit-msg	Enforce Conventional Commits format

Commit Convention

All commits must follow the Conventional Commits specification, enforced by commitizen:

type(scope): description

Types: feat, fix, docs, style, refactor, perf, test, build, ci, chore

Scopes: metrics, runners, generators, storage, schemas, core, llm, guardians, deps

Examples:

feat(metrics): add agentic evaluation metric
fix(toxicity): handle empty batches
docs: update installation guide
test(runners): add integration tests

Claude Code Skills

This project includes Claude Code skills to automate common workflows. If you use Claude Code as your development tool, the following skills are available:

Skill	Command	Description
Commit	/commit	Analyzes staged changes and generates a properly formatted Conventional Commit message
Docs	/docs [type] [name]	Generates Mintlify MDX documentation for new modules (e.g., /docs metric agentic)
Lambda	/fair-forge-lambda	Scaffolds AWS Lambda deployment boilerplate for Fair-Forge modules

License

MIT License - see LICENSE for details.