Hugging Face Daily Papers · August 6, 2026 · 5 min read

Lossless Tensor Compression as Program Synthesis

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Brevis formulates lossless tensor compression as program synthesis, generating compact, executable DSL programs that achieve superior storage reduction and high-speed, bit-exact reconstruction for large model checkpoints.\nPaper: <a href=\"https://arxiv.org/abs/2608.02162\" rel=\"nofollow\">https://arxiv.org/abs/2608.02162</a> Code: <a href=\"https://github.com/jiekeshi/Brevis\" rel=\"nofollow\">https://github.com/jiekeshi/Brevis</a>\n","updatedAt":"2026-08-06T14:33:25.936Z","author":{"_id":"60e3ef01905624186ecba53d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60e3ef01905624186ecba53d/-yaR5Dpcv7AylCLFZpts2.jpeg","fullname":"Jieke SHI","name":"jiekeshi","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.8920623064041138},"editors":["jiekeshi"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/60e3ef01905624186ecba53d/-yaR5Dpcv7AylCLFZpts2.jpeg"],"reactions":[],"isReport":false}},{"id":"6a7539acaef19f4ca4fc6477","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":378,"isUserFollowing":false},"createdAt":"2026-08-07T01:49:32.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Approaching Shannon Bound with Lossless LLM Weight Compression](https://huggingface.co/papers/2606.15789) (2026)\n* [LLM-based Source Code Compression via Thresholded Symbol Ranking](https://huggingface.co/papers/2607.24192) (2026)\n* [Xema: Efficient Diffusion Serving through Fine-Grained Memory Management and Auto-Configuration](https://huggingface.co/papers/2607.11136) (2026)\n* [EVOLVE: Efficient Learned Volume Compression with Variable-Rate Encoding on a Cross-Domain Database](https://huggingface.co/papers/2607.18187) (2026)\n* [SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference](https://huggingface.co/papers/2606.26587) (2026)\n* [Heterogeneity-Aware Microscaling for Efficient Low-Bit LLM Inference](https://huggingface.co/papers/2608.03867) (2026)\n* [FSZ: Breaking the Prediction-Throughput Trade-off in GPU Lossy Compression](https://huggingface.co/papers/2607.15413) (2026)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"This is an automated message from the <a href=\"https://huggingface.co/librarian-bots\">Librarian Bot</a>. I found the following papers similar to this paper. \nThe following papers were recommended by the Semantic Scholar API \n<ul>\n<li><a href=\"https://huggingface.co/papers/2606.15789\">Approaching Shannon Bound with Lossless LLM Weight Compression</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2607.24192\">LLM-based Source Code Compression via Thresholded Symbol Ranking</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2607.11136\">Xema: Efficient Diffusion Serving through Fine-Grained Memory Management and Auto-Configuration</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2607.18187\">EVOLVE: Efficient Learned Volume Compression with Variable-Rate Encoding on a Cross-Domain Database</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2606.26587\">SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2608.03867\">Heterogeneity-Aware Microscaling for Efficient Low-Bit LLM Inference</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2607.15413\">FSZ: Breaking the Prediction-Throughput Trade-off in GPU Lossy Compression</a> (2026)</li>\n</ul>\n Please give a thumbs up to this comment if you found it helpful!\n If you want recommendations for any Paper on Hugging Face checkout <a href=\"https://huggingface.co/spaces/librarian-bots/recommend_similar_papers\">this</a> Space\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: <code>@librarian-bot recommend</code>\n","updatedAt":"2026-08-07T01:49:32.923Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":378,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7270025014877319},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.02162","authors":[{"_id":"6a749853e1228e04b3237fdf","user":{"_id":"60e3ef01905624186ecba53d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60e3ef01905624186ecba53d/-yaR5Dpcv7AylCLFZpts2.jpeg","isPro":false,"fullname":"Jieke SHI","user":"jiekeshi","type":"user","name":"jiekeshi"},"name":"Jieke Shi","status":"claimed_verified","statusLastChangedAt":"2026-08-06T16:45:04.747Z","hidden":false},{"_id":"6a749853e1228e04b3237fe0","name":"Junda He","hidden":false},{"_id":"6a749853e1228e04b3237fe1","name":"Wenjia Jiang","hidden":false},{"_id":"6a749853e1228e04b3237fe2","name":"Weifeng Sun","hidden":false},{"_id":"6a749853e1228e04b3237fe3","name":"Shidong Pan","hidden":false},{"_id":"6a749853e1228e04b3237fe4","name":"Zhensu Sun","hidden":false},{"_id":"6a749853e1228e04b3237fe5","name":"Chengran Yang","hidden":false},{"_id":"6a749853e1228e04b3237fe6","name":"Peixin Zhang","hidden":false},{"_id":"6a749853e1228e04b3237fe7","name":"Yifan Jia","hidden":false},{"_id":"6a749853e1228e04b3237fe8","name":"Zhou Yang","hidden":false},{"_id":"6a749853e1228e04b3237fe9","name":"Thong Hoang","hidden":false},{"_id":"6a749853e1228e04b3237fea","name":"Xiwei Xu","hidden":false},{"_id":"6a749853e1228e04b3237feb","name":"Zhenchang Xing","hidden":false},{"_id":"6a749853e1228e04b3237fec","name":"David Lo","hidden":false}],"publishedAt":"2026-08-03T00:00:00.000Z","submittedOnDailyAt":"2026-08-06T00:00:00.000Z","title":"Lossless Tensor Compression as Program Synthesis","submittedOnDailyBy":{"_id":"60e3ef01905624186ecba53d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60e3ef01905624186ecba53d/-yaR5Dpcv7AylCLFZpts2.jpeg","isPro":false,"fullname":"Jieke SHI","user":"jiekeshi","type":"user","name":"jiekeshi"},"summary":"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 captures recurring tensor structures, such as repeated regions and floating-point fields, through a set of reversible operators. Given a tensor, Brevis synthesizes a self-contained DSL program that reconstructs it bit-exactly. A checkpoint-specific production prior, learned from a small representative sample of tensors, guides a bounded A* search to synthesize compact programs, which can later be executed directly for bit-exact decompression. On 10 public checkpoints spanning language, audio, and image generation models, Brevis reduces 2.13 TB of checkpoint data to 1.41 TB, a 33.93% storage reduction. It produces archives up to 30.87% smaller than those of four general-purpose compressors, including zstd and gzip, and smaller archives than the tensor-specific compressors ZipNN and DFloat11. Under a practical concurrency configuration, Brevis achieves 3.60 GB/s compression and 6.61 GB/s decompression while preserving every source byte.","upvotes":8,"discussionId":"6a749853e1228e04b3237fed","githubRepo":"https://github.com/jiekeshi/Brevis","githubRepoAddedBy":"user","githubStars":1,"organization":{"_id":"6296f43820dc74838613d1ba","name":"SingaporeManagementUniversity","fullname":"Singapore Management University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1654060072621-6296f3e0b493a00f946220de.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"60e3ef01905624186ecba53d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60e3ef01905624186ecba53d/-yaR5Dpcv7AylCLFZpts2.jpeg","isPro":false,"fullname":"Jieke SHI","user":"jiekeshi","type":"user"},{"_id":"65f40d605d44012d77f6251c","avatarUrl":"/avatars/aee2b1c29899111419534427e17aa6f6.svg","isPro":false,"fullname":"Yunbo Lyu","user":"Ra1nbow99","type":"user"},{"_id":"645b0c3ec35da9c7afd95421","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/645b0c3ec35da9c7afd95421/vYBrCDagHsXAo6J2p-uG0.jpeg","isPro":false,"fullname":"Yuling","user":"YerbaPage","type":"user"},{"_id":"6408823b92033c15073b59d5","avatarUrl":"/avatars/24633b955436e638a18a186749f42530.svg","isPro":true,"fullname":"iCSawyer","user":"iCSawyer","type":"user"},{"_id":"6295c00307dbe3912697e983","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6295c00307dbe3912697e983/J6Eyes3Xurv0bvjTbjTjT.jpeg","isPro":false,"fullname":"v587su","user":"zhensuuu","type":"user"},{"_id":"65ce10de8b4adee87192431d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65ce10de8b4adee87192431d/IOg4NGa2_NUxIBvVn0_a-.jpeg","isPro":true,"fullname":"Chen Zhi","user":"hellochenzhi","type":"user"},{"_id":"65195eafb4bfc5d91209e2b1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65195eafb4bfc5d91209e2b1/bUEXt6ZSzqvFKxsd5ml0q.jpeg","isPro":false,"fullname":"Xiuwei Shang","user":"Sxxxw","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6296f43820dc74838613d1ba","name":"SingaporeManagementUniversity","fullname":"Singapore Management University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1654060072621-6296f3e0b493a00f946220de.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.02162.md","query":{}}">

Papers

arxiv:2608.02162

Lossless Tensor Compression as Program Synthesis

Published on Aug 3

· Submitted by

Jieke SHI on Aug 6

Singapore Management University

Upvote

Authors:

Jieke Shi ,

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 captures recurring tensor structures, such as repeated regions and floating-point fields, through a set of reversible operators. Given a tensor, Brevis synthesizes a self-contained DSL program that reconstructs it bit-exactly. A checkpoint-specific production prior, learned from a small representative sample of tensors, guides a bounded A* search to synthesize compact programs, which can later be executed directly for bit-exact decompression. On 10 public checkpoints spanning language, audio, and image generation models, Brevis reduces 2.13 TB of checkpoint data to 1.41 TB, a 33.93% storage reduction. It produces archives up to 30.87% smaller than those of four general-purpose compressors, including zstd and gzip, and smaller archives than the tensor-specific compressors ZipNN and DFloat11. Under a practical concurrency configuration, Brevis achieves 3.60 GB/s compression and 6.61 GB/s decompression while preserving every source byte.

View arXiv page View PDF GitHub 1 Add to collection