Decode-Time Grammars: Constrained LLM Generation over a Refinement Order of Grammar Fragments
2026-07-22T08:52:01Z•0cbd48f3758f27ae77c9e9e348439141286ec58c1851130a818eead7211062d7
LLM safetyLLVMMLIRSQL safetyVirtualSetbuild authorizationcompiler securityconstrained decodingdecode-time grammarsformal methodsghost referencesgrounded generationguarded executionhardware verificationhigh-level synthesislink receiptspattern-unificationruntime environment typingschema-derived priorsweak-memory verification
What happened
Collection of PL/AI research (arXiv 22 Jul 2026) focused on safer, more precise program generation and verification. Highlights include: decode-time grammars (gproj) that instantiate grammar fragments from a runtime environment to eliminate “ghost” references during LLM code generation; VirtualSet, a typed ontology/world interface that turns invalid SQL-like operations into pre-execution type errors and supports guarded simulated writes; a checked build-authorized boundary for compiler rewrites (link receipts, fail-closed authorization) to constrain optimization authority; schema-derived, MLIR
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_pl
- Record identifier
- 0cbd48f3758f27ae77c9e9e348439141286ec58c1851130a818eead7211062d7
- Enrichment time
- 2026-07-22T08:52:01Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.