# huggingface/search-and-learn

Recipes to scale inference-time compute of open models

Repository: https://github.com/huggingface/search-and-learn
Canonical: https://ross.abutalabs.com/products/search-and-learn
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-05-26T10:33:21+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 84, release rhythm 35, longevity 45
- inputs: {"age_days": 632, "days_push": 99, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1131, forks 131 (observed 2026-08-28T04:03:42.450651+00:00)

## What it is
Search and Learn is a collection of Python scripts and YAML recipes from Hugging Face for scaling inference-time (test-time) compute of open LLMs using search algorithms like Best-of-N, beam search, and Diverse Verifier Tree Search. It accompanies a blog post and lets users replicate test-time compute scaling results.

## Use cases
- scale test-time compute for open LLMs
- run beam search over LLM generations
- replicate o1-style test-time scaling results
- apply best-of-N sampling with verifiers
- run diverse verifier tree search on math problems
- learn inference-time compute scaling techniques

## When to choose
- you want to improve open-model reasoning by spending more compute at inference
- you want to reproduce published test-time compute scaling experiments
- you need ready-made recipes for search strategies like beam search or DVTS

## When to avoid
- you need a production inference server or API
- you want training or fine-tuning recipes rather than inference-time methods
- you need a maintained library with stable APIs rather than research scripts

## Facets
- artifact type: learning-resource
- maturity: active
- function: llm-inference, machine-learning, benchmarking
- domain: large-language-models, machine-learning, artificial-intelligence
- platform: python
- tags: test-time-compute, search-algorithms, beam-search, best-of-n, tree-search, recipes, open-models, gpu, linux

## Member repositories
- huggingface/search-and-learn (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:42.450651+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:38:02.580324+00:00, confidence not recorded.
  - readme: https://github.com/huggingface/search-and-learn (fetched 2026-08-28T04:03:42.450651+00:00, sha 726353941b62)
- Data as of 2026-08-30T08:39:29.467469+00:00.