step-level-verification-structured-quantitative-stem-reasoning-training-7b-process CLM-0643mechanismsingle paperReviewed by AI
For step-level verification of structured quantitative STEM reasoning, training a 7B process reward model only on hard negatives that are constructed to pass a deterministic symbolic verifier (swapped formula or unit-matched variable, with values recomputed so the math balances), and running the verifier as a hard filter before the PRM scores candidates, detects flawed steps better than monolithic PRMs that must judge arithmetic and logic together.
Capability: Spotting an error in a given artifact
Observed on
Problems expressible as sequences of deterministic calculations or algebraic manipulations with units; verifier coverage limited to the schema's ~120 primitives; excludes qualitati.
Sources
- supportsNeuro-symbolic PRM: Enhancing Scientific Reasoning via Structured Traces and Symbolic VerificationValidated by: ProcessBench average F1 74.2 for verifier+CSP-PRM vs 73.5 for Qwen2.5-Math-PRM-7B and 68.8 for the same CSP-PRM without the verifier; PRMBench soundness/sensitivity gains; Best-of-8 downstream average 51.5 vs 49.4 best baseline; perturbation ablation with identical positives gave 75.4 joint vs 71.2 logic-only and 70.9 operand-only.. Measured on ProcessBench, PRMBench, Best-of-8 decoding and science benchmarks with one 7B backbone; verifier and CSP data are ablated separately and the two perturbation policies are isolated with identical positives. [truncated]
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- On natural-language logical reasoning benchmarks, auto-formalizing each reasoning step into SMT-LIB and regenerating from the first step the solver cannot verify raises both the share of solver-verified steps and final-answer accuracy over plain chain-of-thought, and greedy regeneration at the failure point beats beam search or MCTS on step verifiability.Carrying a chain of inference through to its conclusion · reviewed by AI
- On human-annotated step-level verification of frontier-model proofs to recent IMO and Putnam problems, most of 29 generative critics and process reward models do little better than chance at finding the first wrong step, a leading math PRM (Qwen2.5-Math-PRM-72B) reaches only about 37% balanced F1, and beyond a few standouts open-source verifiers lag closed models; verification is harder than generation for the same frontier models.Spotting an error in a given artifact · reviewed by AI
- Outside mathematics, the step-level error detection of both process reward models and general LLMs is markedly weaker than on math benchmarks, and the two fail differently: PRMs miss knowledge-based errors while LLMs miss computational errors, across science and logic domains.Spotting an error in a given artifact · reviewed by AI
- When training small instruction-tuned models with reinforcement learning plus factual supervision, routing each atomic fact's verification score only to the tokens that produced it, and down-weighting verifier judgements that do not change when their key evidence is removed, improves factuality benchmark scores over trajectory-level or reasoning- step-level factual rewards.Checking claims against evidence · reviewed by AI
- For a same-family pair of reasoning and instruction-tuned 4B models, learning one interpolation ratio per module (attention block, FFN block, embedding, final norm) with a reward that scores correctness and ranks correct responses by length cuts average generated tokens by about 27% relative to the reasoning model while holding average accuracy, and beats fixed global-ratio merging and CMA-ES search over the same ratios.Cutting to the chase · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: A PRM trained on ordinary mixed negatives, with the same verifier-first pruning at inference and matched compute, matching or beating the constraint-preserving-hard-negative PRM on step-level F1.