Elamonica - Test-Time Compute Optimization Framework

A production-grade framework for optimizing LLM inference through intelligent test-time compute allocation.

Overview

Elamonica implements state-of-the-art test-time compute optimization strategies from recent research, enabling developers to achieve better inference quality through strategic compute allocation.

Key Features

Community Edition (Open Source)

• Best-of-N sampling with configurable parameters
• Sequential revision strategies
• Beam search optimization
• Command-line interface
• Comprehensive benchmarking tools

Pro Edition

• Adaptive compute allocation
• Process Reward Model (PRM) integration
• Web-based dashboard
• REST API server
• License management system

Enterprise Edition

• White-label customization
• SSO integration
• Custom PRM training pipeline
• Multi-tenant architecture
• Kubernetes operator

Architecture

Elamonica is built on three core optimization strategies:

1. Parallel Strategies: Generate multiple candidate responses and select the best
2. Sequential Strategies: Iteratively refine responses through multiple passes
3. Search-Based Strategies: Use guided search with reward models

Installation

Community Edition

pip install elamonica

From Source

git clone https://github.com/AntonioVFranco/elamonica.git
cd elamonica
pip install -e community/

Quick Start

from elamonica import OptimizationPipeline

# Initialize pipeline with best-of-N strategy
pipeline = OptimizationPipeline(
    model="deepseek-ai/deepseek-r1-distill-qwen-32b",
    strategy="best_of_n",
    n_samples=5
)

# Optimize inference
result = pipeline.optimize(
    prompt="Solve the following problem: ...",
    max_compute_budget=100
)

print(result.best_response)

Documentation

• Installation Guide
• API Reference
• Architecture Overview
• Contributing Guidelines

License

• Community Edition: Apache 2.0
• Pro Edition: Commercial License
• Enterprise Edition: Commercial License

See LICENSE for details.

Citation

@software{elamonica2025,
  title={Elamonica: Test-Time Compute Optimization Framework},
  author={Antonio Silva},
  year={2025},
  url={https://github.com/AntonioVFranco/elamonica}
}

Support

Feel free to contact me via email for any needs: contact@antoniovfranco.com

• Documentation: docs.elamonica
• Issues: GitHub Issues
• Discussions: GitHub Discussions