Source · Tier B · PreprintSource · Bit-Exact AI Inference Verification Without Performance Tradeoffs
Bit-Exact AI Inference Verification Without Performance Tradeoffs
N. Cankaya. 2026. ICML 2026 Workshop on Technical AI Governance Research.
| Original | https://arxiv.org/abs/2606.00279 |
|---|---|
| DOI | 10.48550/arXiv.2606.00279 |
| arXiv | 2606.00279 |
| Version | arXiv v1 29 May 2026; v2 5 June 2026. The arXiv comments field reads "Best paper award, ICML 2026 TAIGR workshop"; the paper is listed in the workshop's poster session (https://icml.cc/virtual/2026/workshop/54084). Checked 2026-09-24. |
| Accessed | 2026-09-25 |
| Imported from | hodgkins-ai-verification-papers@c71e59ff0e8e |
| Note | Listed under "Inference verification" in the Hodgkins bibliography (CC BY 4.0). |
Cited by
- Deterministic and bit-exact inferenceMechanism
- Sampled inference recomputationMechanism
- Batch-invariant inference kernels (Thinking Machines)Implementation
- DiFR (Divergence From Reference)Implementation
- TOPLOCImplementation
- The declared model is the one being servedClaim
- This compute runs inference, not trainingClaim
- Model weights have not left the facilityClaim
- Numerical nondeterminismConcept
- Machine Intelligence Research InstituteOrganization