ReasonFlux
[NeurIPS 2025 Spotlight] LLM post-training suite — featuring ReasonFlux, ReasonFlux-PRM, and ReasonFlux-Coder.
- 545
- Python
- 14 mo ago
[NeurIPS 2025 Spotlight] LLM post-training suite — featuring ReasonFlux, ReasonFlux-PRM, and ReasonFlux-Coder.
Fixing GRPO training collapse in long-horizon multi-tool agents. A lightweight PRM-Lite + LATA joint approach achieves +37% over vanilla GRPO on τ-bench airline (50-task, multi-turn).
[AAAI 2026] Official codebase for "GenPRM: Scaling Test-Time Compute of Process Reward Models via Generative Reasoning".
[ACL 2026 SAC Highlight Award] Can We Predict Before Executing Machine Learning Agents?
This repository contains the official implementation of Error Typing for Smarter Rewards: Improving Process Reward Models with Error-Aware Hierarchical Supervision.
Step-level preference data construction toolkit for code-agent process reward models
GenPark Reasoning MCTS Step Verity Evaluator - Implements Monte Carlo Tree Search (MCTS) Step-Level Process Reward Modeling (PRM) & Hallucination Branch Pruning.
GenPark Reasoning Majority Voting Consensus Engine - Implements Self-Consistency Best-of-N temperature resampler with semantic deduplication and confidence-weighted voting.
GenPark Reasoning Formal Symbolic Proof Checker - Zero-dependency propositional logic assertion and formal symbolic deduction validator.
GenPark Reasoning Budget Test-Time Compute Allocator - Dynamically schedules thinking token budgets, reflection passes, and temperature annealing based on problem entropy.