First usable release — package name now matches imports.
Install
pip install thinkbooster
Usage
from thinkbooster.strategies.strategy_beam_search import StrategyBeamSearch
from thinkbooster.evaluation.grader import math_equal
from thinkbooster.evaluation.parser import extract_answer
Highlights
- 9 test-time scaling strategies (beam search, best-of-N, self-consistency, and more)
- vLLM, HuggingFace, and OpenAI-compatible backends
- Evaluation pipeline for MATH-500, GSM8K, OlympiadBench, AIME, HumanEval+, MBPP+
- Visual debugger and OpenAI-compatible REST API
- Verified: MATH-500 baseline with Qwen2.5-Math-7B-Instruct — 83.2% exact_match