Reward-Weighted On-Policy Distillation with an Open Property-Equivalence Verifier for NL-to-SVA Generation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zou, Qingyun, Li, Yingze, Liu, Tianen, He, Bingsheng, Wong, Weng-Fai
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914563379691520
author Zou, Qingyun
Li, Yingze
Liu, Tianen
He, Bingsheng
Wong, Weng-Fai
author_facet Zou, Qingyun
Li, Yingze
Liu, Tianen
He, Bingsheng
Wong, Weng-Fai
contents LLM-based generation of SystemVerilog Assertions (SVA) is often reported as nearing saturation, with the strongest specialized model reaching ${\sim}76\%$ accuracy on NL2SVA-Human. We show that this aggregate hides a temporal gap: models that appear strong overall still collapse to a few implication templates on bounded-delay and liveness specifications. The core issue is that the dominant recipe, supervised fine-tuning on NL/SVA pairs, optimizes token-level mimicry rather than the \emph{property equivalence} that defines SVA correctness. We introduce \emph{Reward-Weighted On-Policy Distillation} (RWOPD), an on-policy distillation method that samples student rollouts, scores them with an open SymbiYosys+Z3 Property-Equivalence Checker (PEC), and applies a verifier-reward-weighted forward-KL gradient from a frozen 14B teacher on verifier-passable rollouts. This keeps the supervision dense at every response token while grounding both selection and loss weight in property-equivalent behavior. RWOPD distills CodeV-SVA-14B into a Qwen2.5-Coder-7B-Instruct student that sets a new state of the art on NL2SVA-Human and NL2SVA-Machine across pass@1, pass@5, and pass@10, surpassing both specialized prior SOTA models and 671B general-purpose baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2605_13501
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Reward-Weighted On-Policy Distillation with an Open Property-Equivalence Verifier for NL-to-SVA Generation
Zou, Qingyun
Li, Yingze
Liu, Tianen
He, Bingsheng
Wong, Weng-Fai
Hardware Architecture
Machine Learning
LLM-based generation of SystemVerilog Assertions (SVA) is often reported as nearing saturation, with the strongest specialized model reaching ${\sim}76\%$ accuracy on NL2SVA-Human. We show that this aggregate hides a temporal gap: models that appear strong overall still collapse to a few implication templates on bounded-delay and liveness specifications. The core issue is that the dominant recipe, supervised fine-tuning on NL/SVA pairs, optimizes token-level mimicry rather than the \emph{property equivalence} that defines SVA correctness. We introduce \emph{Reward-Weighted On-Policy Distillation} (RWOPD), an on-policy distillation method that samples student rollouts, scores them with an open SymbiYosys+Z3 Property-Equivalence Checker (PEC), and applies a verifier-reward-weighted forward-KL gradient from a frozen 14B teacher on verifier-passable rollouts. This keeps the supervision dense at every response token while grounding both selection and loss weight in property-equivalent behavior. RWOPD distills CodeV-SVA-14B into a Qwen2.5-Coder-7B-Instruct student that sets a new state of the art on NL2SVA-Human and NL2SVA-Machine across pass@1, pass@5, and pass@10, surpassing both specialized prior SOTA models and 671B general-purpose baselines.
title Reward-Weighted On-Policy Distillation with an Open Property-Equivalence Verifier for NL-to-SVA Generation
topic Hardware Architecture
Machine Learning
url https://arxiv.org/abs/2605.13501