Skip to main content

A Python library for inference-time scaling LLMs

Project description

its_hub: A Python library for inference-time scaling LLMs

Example: Using the particle filtering from [1] for inference-time scaling

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

# NOTE launched via `CUDA_VISIBLE_DEVICES=0 vllm serve Qwen/Qwen2.5-Math-1.5B-Instruct --dtype float16`
lm = OpenAICompatibleLanguageModel(
    endpoint="http://0.0.0.0:8000/v1", 
    api_key="NO_API_KEY", 
    model_name="Qwen/Qwen2.5-Math-1.5B-Instruct", 
    system_prompt=SAL_STEP_BY_STEP_SYSTEM_PROMPT, 
)
prompt = r"Let $a$ be a positive real number such that all the roots of \[x^3 + ax^2 + ax + 1 = 0\]are real. Find the smallest possible value of $a.$" # question from MATH500
budget = 8

sg = StepGeneration(r"\n\n", 32, r"\boxed")
prm = LocalVllmProcessRewardModel(model_name="Qwen/Qwen2.5-Math-PRM-7B", device="cuda:1")
scaling_alg = ParticleFiltering(sg, prm)

scaling_alg.infer(lm, prompt, budget, show_progress=True) # => gives output

[1]: Isha Puri, Shivchander Sudalairaj, Guangxuan Xu, Kai Xu, Akash Srivastava. “A Probabilistic Inference Approach to Inference-Time Scaling of LLMs using Particle-Based Monte Carlo Methods”, 2025.

Installation

Latest release from PyPI

pip install its_hub

Latest main branch

pip install git+https://github.com/Red-Hat-AI-Innovation-Team/its_hub.git

Benchmark

There is a script at scripts/benchmark.py that can be used to benchmark inference-time scaling algorithms. The CLI of the script is self-contained so the usage can be checked via python scripts/benchmark.py --help. Example command:

python scripts/benchmark.py --benchmark aime-2024 --model_name Qwen/Qwen2.5-Math-1.5B-Instruct --alg particle-filtering --rm_device cuda:1 --endpoint http://0.0.0.0:8000/v1 --shuffle_seed 1110 --does_eval --budgets 1,2,4,8,16,32,64 --rm_agg_method model

Development

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

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

its_hub-0.1.0a1.tar.gz (9.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

its_hub-0.1.0a1-py3-none-any.whl (9.7 kB view details)

Uploaded Python 3

File details

Details for the file its_hub-0.1.0a1.tar.gz.

File metadata

  • Download URL: its_hub-0.1.0a1.tar.gz
  • Upload date:
  • Size: 9.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.12.9

File hashes

Hashes for its_hub-0.1.0a1.tar.gz
Algorithm Hash digest
SHA256 0e6eca18616e0fb58ce262b08912c01869fb8195c817c894e23022b42c687776
MD5 11abf96bd4443ac60504aa1951698299
BLAKE2b-256 d73398e1536c93c393315174416eb162fd6eba9f9eeefe74ebf5710d520eaa68

See more details on using hashes here.

Provenance

The following attestation bundles were made for its_hub-0.1.0a1.tar.gz:

Publisher: release.yaml on Red-Hat-AI-Innovation-Team/its_hub

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file its_hub-0.1.0a1-py3-none-any.whl.

File metadata

  • Download URL: its_hub-0.1.0a1-py3-none-any.whl
  • Upload date:
  • Size: 9.7 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.12.9

File hashes

Hashes for its_hub-0.1.0a1-py3-none-any.whl
Algorithm Hash digest
SHA256 8513dab0828e1aebc1971464bfb77ae5c867f70608c091ce45cb6371a2a41cb2
MD5 8f8caaeb4d7b8a2424fd67d6acefa6f0
BLAKE2b-256 f4fb44eb9c9519ec3d984556a6526ad679fd88b107ad59e4664f314805e8c0a4

See more details on using hashes here.

Provenance

The following attestation bundles were made for its_hub-0.1.0a1-py3-none-any.whl:

Publisher: release.yaml on Red-Hat-AI-Innovation-Team/its_hub

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page