Conditional Program 压缩法：DeepSeek 记忆表存储从 203GB 降到 51GB
N-Gram Memories at Two Bits: Quantizing Lookup Tables in Large Language Models
arXiv 上的新压缩论文，把 DeepSeek 的 203GB 查找表压到 51GB，做量化或本地部署的可以看看思路。
论文提出 Conditional Program（CP）压缩表示，只重建输入所请求的行，而不必保留完整嵌入表。每行用短编码加局部 scale 表示，共享一个跨表重建程序。在三个训练好的 n-gram 记忆家族上，2-bit 编码的输出 KL 相比 2-bit round-to-nearest 降低 41%-63%。在相同存储预算下，共享 trellis 优于测试的拟合 codebook 和低秩残差变体。对 DeepSeek 而言，表存储从 203GB 降至 51GB。
N-Gram Memories at Two Bits: Quantizing Lookup Tables in Large Language Models
Conditional memory expands language-model capacity through selective lookup, but the full embedding table must remain available for every possible input. We introduce Conditional Program (CP), a compressed representation that reconstructs only the rows an input requests. CP represents each row by a short code and a local scale, with a reconstruction program shared across the table. We instantiate this representation with a Gaussian trellis, norm restoration and independently decodable row records. CP's two-bit code lowers output KL by 41--63\% relative to 2-bit round-to-nearest. Analysis across three trained n-gram memory families shows that Gaussian-like coordinate statistics coexist with learned relationships among rows; at matched storage budgets, the shared trellis outperforms the fitted codebooks and low-rank residual variants we test. For DeepSeek, it reduces table storage from 203GB to 51GB, leaving room for larger workloads when the table is kept on GPU.
