Advancements in Large Language Model Efficiency: A Literature Review on 1-bit Quantization

Author(s)	Lalitha Shree C P, Nethravathi B
Country	India
Abstract	Large Language Models suffer from most of the challenges regarding computational cost, memory, and energy consumption, which makes their scaling difficult. BitNet b1.58 addresses these issues by introducing a novel 1.58-bit quantization with ternary weights {-1, 0, 1}. This can achieve performance comparable to FP16 models with much lower resource requirements. BitNet b1.58 provides 2.71x faster inference and 3.55x less memory usage than FP16 baselines. Its ternary weights allow for efficient feature filtering, making it a versatile choice for many AI applications. This makes it a standout solution, balancing high performance with resource efficiency. While BitNet b1.58 has its 1-bit mover heads that make it affordable for edge and mobile devices, it also allows longer sequences. This will mark further developments toward making AI scalable and resource-aware for various applications.
Keywords	LLMs, 1 bit Quantization, BitNet, BitNet b1.58
Field	Engineering
Published In	Volume 6, Issue 6, November-December 2024
Published On	2024-12-10
Cite This	Advancements in Large Language Model Efficiency: A Literature Review on 1-bit Quantization - Lalitha Shree C P, Nethravathi B - IJFMR Volume 6, Issue 6, November-December 2024. DOI 10.36948/ijfmr.2024.v06i06.32856
DOI	https://doi.org/10.36948/ijfmr.2024.v06i06.32856
Short DOI	https://doi.org/g8vgf5

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9898-5948-55	editor@ijfmr.com

International Journal For Multidisciplinary Research