Lossless and Near-Lossless Compression for Foundation Models
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:B7BHG44Orecord.jsonopen to challenge →
read the original abstract
With the growth of model sizes and scale of their deployment, their sheer size burdens the infrastructure requiring more network and more storage to accommodate these. While there is a vast literature about reducing model sizes, we investigate a more traditional type of compression -- one that compresses the model to a smaller form and is coupled with a decompression algorithm that returns it to its original size -- namely lossless compression. Somewhat surprisingly, we show that such lossless compression can gain significant network and storage reduction on popular models, at times reducing over $50\%$ of the model size. We investigate the source of model compressibility, introduce compression variants tailored for models and categorize models to compressibility groups. We also introduce a tunable lossy compression technique that can further reduce size even on the less compressible models with little to no effect on the model accuracy. We estimate that these methods could save over an ExaByte per month of network traffic downloaded from a large model hub like HuggingFace.
This paper has not been read by Pith yet.
Forward citations
Cited by 3 Pith papers
-
Holmes: A Benchmark to Assess the Linguistic Competence of Language Models
Holmes is a probing benchmark compiling over 200 datasets from 270 studies to evaluate linguistic competence across syntax, morphology, semantics, reasoning, and discourse in more than 50 language models.
-
OpenZL: Using Graphs to Compress Smaller and Faster
OpenZL uses a directed acyclic graph of modular codecs to enable rapid creation of application-specific compressors that deliver better ratios and speeds than general-purpose tools while remaining competitive with dee...
-
Smaller Models, Unexpected Costs: Trade-offs in LLM Quantization for Automated Program Repair
Empirical evaluation of 13 quantization configurations on 6 LLMs for APR shows reduced memory (up to 85%) but increased inference time/energy, different repaired problem sets with little overlap, and 48% of configs st...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.