让120亿参数模型零代价复用已验证解题记忆，永久准确无误。
A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever
- 模型冻结，仅靠持久记忆库复用已验证答案，无需生成
- 9个题型180例全对，零生成令牌，结果比特精确可复现
- 适合需要高可靠、低延迟推理的工业级应用，如正式证明系统
当前语言模型优化依赖重复训练：消耗巨大算力，每次生成新模型且输出不可控。本文反其道而行之：模型保持冻结，仅构建一个持续积累的已验证解决方案记忆库。一旦某类问题被独立验证（不依赖答案键），后续同类问题均以零生成令牌、比特精确、确定性方式回答。在涵盖九类问题的180个新实例上，四个厂商的四种架构（密集与专家混合）均实现180/180正确率，每答一问零生成令牌，推理能力与参数规模解耦。负向对照实验表明，清空记忆库后系统无法解决任何问题。该机制同样适用于开放推理：四模型在88个案例中一致通过一致性门控，经机器检查形式证明，推理方法迁移成功率77/80。记忆检索耗时仅1.4微秒；完整复用耗时6-23毫秒，能耗36毫瓦。在4,500项已验证条目中，近似相似检索错误率达94.3%，而精确地址访问零错误。该记忆库还作为可移动上下文窗口，支持600万令牌规模，单块46GB GPU即可维持，远超vLLM的30,399令牌和SGLang的32,000令牌限制。在公开基准上，前沿模型仍显著优于所有12B模型的原始推理能力；但在本系统已验证并解决的问题上，对比反转：前沿API需为每查询付出一次生成开销，永无止境，而本系统复用成本为零，返回完全一致的比特结果。配套公共测试平台已开放：https://corbenic-galahad-bench.hf.space
原文摘要 · Abstract (English)
Improving a language model today means retraining it: enormous compute, a new opaque model each cycle, non-deterministic output. We take the opposite path: the model stays frozen, and a persistent memory of verified solutions grows beside it. Once a problem family is solved and has passed an independent verification step that never consults the answer key, every new instance of that family is answered at zero generation tokens, bit-exact, deterministically. Across 180 fresh instances spanning nine problem families, four architectures from four vendors - dense and mixture-of-experts - each score 180/180 at zero generation tokens per answer: execution-bound capability decoupled from parameter scaling. A negative control attributes the capability fully to the memory: emptied, it solves nothing. The same verify-before-store contract holds for open-ended reasoning: 88/88 consistency-gated acceptances across all four models, machine-checked formal proof, and reasoning-method transfer at 77/80. Memory selection takes 1.4 microseconds; a full reuse completes in 6-23 ms at 36 mWh. Approximate similarity retrieval selects the wrong item 94.3% of the time on a 4,500-item verified store where exact addressing makes zero errors. The store also serves as working context at a scale no shipped engine matches: a 6,000,000-token movable window on a single 46 GB GPU at flat memory, where vLLM stops at 30,399 tokens and SGLang silently truncates past 32,000. On published benchmarks, frontier models remain far ahead of any 12B at raw from-scratch reasoning; on everything this system has solved and verified, the comparison inverts: a frontier API call pays a fresh generation pass on every query, forever, while verified reuse costs zero tokens and returns the identical bits every time. A public testbench with free, rate-limited access accompanies this report: https://corbenic-galahad-bench.hf.space
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。