dorsal/arxiv
View SchemaSingle-Stage Huffman Encoder for ML Compression
| Authors | Aditya Agrawal, Albert Magyar, Hiteshwar Eswaraiah, Patrick Sheridan, Pradeep Janedula, Ravi Krishnan Venkatesan, Krishna Nair, Ravi Iyer |
|---|---|
| Categories | |
| ArXiv ID | 2601.10673vv1 |
| URL | https://arxiv.org/abs/2601.10673 |
| License | http://creativecommons.org/licenses/by/4.0/ |
Abstract
Training and serving Large Language Models (LLMs) require partitioning data across multiple accelerators, where collective operations are frequently bottlenecked by network bandwidth. Lossless compression using Huffman codes is an effective way to alleviate the issue, however, its three-stage design requiring on-the-fly frequency analysis, codebook generation and transmission of codebook along with data introduces computational, latency and data overheads which are prohibitive for latency-sensitive scenarios such as die-to-die communication. This paper proposes a single-stage Huffman encoder that eliminates these overheads by using fixed codebooks derived from the average probability distribution of previous data batches. Through our analysis of the Gemma 2B model, we demonstrate that tensors exhibit high statistical similarity across layers and shards. Using this approach we achieve compression within 0.5% of per-shard Huffman coding and within 1% of the ideal Shannon compressibility, enabling efficient on-the-fly compression.
{
"annotation_id": "0416300a-414a-4e6f-b6af-bfbc2e15837b",
"date_created": "2026-02-17T05:53:26.424000Z",
"date_modified": "2026-02-17T05:53:26.424000Z",
"file_hash": "a10b3dbe1fd1a866f4970783ee03770dedc59121a16fa58070137cd137902ab3",
"private": false,
"record": {
"abstract": "Training and serving Large Language Models (LLMs) require partitioning data across multiple accelerators, where collective operations are frequently bottlenecked by network bandwidth. Lossless compression using Huffman codes is an effective way to alleviate the issue, however, its three-stage design requiring on-the-fly frequency analysis, codebook generation and transmission of codebook along with data introduces computational, latency and data overheads which are prohibitive for latency-sensitive scenarios such as die-to-die communication. This paper proposes a single-stage Huffman encoder that eliminates these overheads by using fixed codebooks derived from the average probability distribution of previous data batches. Through our analysis of the Gemma 2B model, we demonstrate that tensors exhibit high statistical similarity across layers and shards. Using this approach we achieve compression within 0.5% of per-shard Huffman coding and within 1% of the ideal Shannon compressibility, enabling efficient on-the-fly compression.",
"arxiv_id": "2601.10673",
"authors": [
"Aditya Agrawal",
"Albert Magyar",
"Hiteshwar Eswaraiah",
"Patrick Sheridan",
"Pradeep Janedula",
"Ravi Krishnan Venkatesan",
"Krishna Nair",
"Ravi Iyer"
],
"categories": [
"cs.LG"
],
"license": "http://creativecommons.org/licenses/by/4.0/",
"title": "Single-Stage Huffman Encoder for ML Compression",
"url": "https://arxiv.org/abs/2601.10673",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "0769bbf9-8647-4844-b9e3-ceb6ee53fcf7",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}