Brevis：将无损张量压缩转化为程序合成
原标题：Lossless Tensor Compression as Program Synthesis
AI 摘要
Brevis 是一种新型无损张量压缩方法，将压缩问题转化为程序合成任务，通过类型化 DSL 和可逆算子捕捉张量结构，并利用生产先验引导 A* 搜索生成可执行程序。在 10 个公开检查点上，Brevis 将 2.13 TB 数据压缩至 1.41 TB，存储减少 33.93%，压缩率优于 zstd、gzip 及 ZipNN 等专用压缩器，同时实现 3.60 GB/s 压缩和 6.61 GB/s 解压速度。
正文节选
Lossless Tensor Compression as Program Synthesis Abstract Model checkpoints are growing in both number and size, which makes archival, transfer, and deployment increasingly costly. General-purpose compressors can reduce storage requirements but ignore tensor structure, whereas existing tensor-specific compressors rely on fixed and format-specific pipelines. We present Brevis, which formulates lossless tensor compression as program synthesis. We design a typed domain-specific language (DSL) that