195 kB
8 files
Updated 8 days ago
README.md

Quantitative LLM Reasoning Vault

License Hub Type Interactive Arena Process Reward Model Benchmarks

📌 Overview

This vault contains the production-grade research codebase, evaluation harnesses, synthetic CoT generation pipelines, and cryptographic manifests for Process Reward Supervision and Quantitative Reasoning LLM Benchmarking.

📁 Manifest & File Structure

File Type Purpose
evaluator.py Python CLI & Library Benchmarking suite scoring reasoning chains against ground truth
prm_verifier.py Python Class Process Reward Model step scoring & fallacy pinpointer
synthetic_cot_generator.py Python Script Procedural generator of multi-step stochastic calculus derivations
benchmark_suite.json Dataset JSON High-density benchmark test split with ground truth CoTs
arena_kernel.js JavaScript Kernel In-browser & Node.js verification engine for step evaluation
quickstart.ipynb Jupyter Notebook Interactive tutorial demonstrating PRM step scoring
MANIFEST.json Cryptographic Ledger SHA256 hashes and byte lengths for all vault assets
README.md Markdown Documentation Reproduction instructions and architectural specification

🚀 Quick Execution

# Run benchmark evaluation
python evaluator.py --benchmarks benchmark_suite.json --samples 25

# Generate synthetic CoT traces
python synthetic_cot_generator.py

🔗 Integrated Ecosystem

Total size
195 kB
Files
8
Last updated
Nov 14
Pre-warmed CDN
US EU US EU

Contributors