its-hub: A Python library for inference-time scaling

its_hub is a Python library for inference-time scaling of LLMs, focusing on mathematical reasoning tasks.

📚 Documentation

For comprehensive documentation, including installation guides, tutorials, and API reference, visit:

https://ai-innovation.team/its_hub

Installation

Choose the installation option based on which algorithms you need:

# Core installation - includes:
#   - Best-of-N with LLM Judge
#   - Self-Consistency
#   - OpenAI-compatible language models
pip install its_hub

# Process Reward Model installation - adds:
#   - Particle Filtering Algorithms
#   - Beam Search
#   - LocalVllmProcessRewardModel
#   - Required for step-by-step reasoning with process reward models
pip install its_hub[prm]

# Development installation
git clone https://github.com/Red-Hat-AI-Innovation-Team/its_hub.git
cd its_hub
pip install -e ".[dev]"

Quick Start

Example 1: Best-of-N with LLM Judge

Installation required: pip install its_hub (core)

Use Best-of-N algorithm with an LLM judge for response selection - works with any OpenAI-compatible API:

from its_hub.lms import OpenAICompatibleLanguageModel
from its_hub.algorithms import BestOfN
from its_hub.integration.reward_hub import LLMJudgeRewardModel

# Initialize language model
lm = OpenAICompatibleLanguageModel(
    endpoint="https://api.openai.com/v1",
    api_key="your-api-key",
    model_name="gpt-4o-mini",
)

# Set up LLM judge for scoring
judge = LLMJudgeRewardModel(
    model="gpt-4o-mini",
    criterion="overall_quality",
    judge_type="groupwise",
    api_key="your-api-key",
)
scaling_alg = BestOfN(judge)

# Generate multiple responses and select the best
result = scaling_alg.infer(lm, "Explain quantum entanglement in simple terms", budget=4)
print(result)

Example 2: Particle Filtering with Process Reward Model

Installation required: pip install its_hub[prm]

Use Particle Filtering for step-by-step reasoning with process reward models:

from its_hub.utils import SAL_STEP_BY_STEP_SYSTEM_PROMPT
from its_hub.lms import OpenAICompatibleLanguageModel, StepGeneration
from its_hub.algorithms import ParticleFiltering
from its_hub.integration.reward_hub import LocalVllmProcessRewardModel

# Initialize language model (requires vLLM server running)
lm = OpenAICompatibleLanguageModel(
    endpoint="http://localhost:8100/v1",
    api_key="NO_API_KEY",
    model_name="Qwen/Qwen2.5-Math-1.5B-Instruct",
    system_prompt=SAL_STEP_BY_STEP_SYSTEM_PROMPT,
)

# Set up step generation and process reward model
sg = StepGeneration(step_token="\n\n", max_steps=32, stop_token=r"\boxed")
prm = LocalVllmProcessRewardModel(
    model_name="Qwen/Qwen2.5-Math-PRM-7B",
    device="cuda:0",
    aggregation_method="prod"
)
scaling_alg = ParticleFiltering(sg, prm)

# Solve with step-by-step reasoning
result = scaling_alg.infer(lm, "Solve x^2 + 5x + 6 = 0", budget=8)
print(result)

Key Features

• 🔬 Multiple Algorithms: Particle Filtering, Best-of-N, Beam Search, Self-Consistency
• 🚀 OpenAI-Compatible API: Easy integration with existing applications
• 🧮 Math-Optimized: Built for mathematical reasoning with specialized prompts
• 📊 Benchmarking Tools: Compare algorithms on MATH500 and AIME-2024 datasets
• ⚡ Async Support: Concurrent generation with limits and error handling

For detailed documentation, visit: https://ai-innovation.team/its_hub