Research map / Architectures and efficiency / Quantization
MemeBQ:Memory Efficient Binary Quantization of LLMs
Yuanhui Wang, Kunlong Liu, Minnan Pei, Zhangming Li, Peisong Wang, Qinghao Hu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add group bitmap search (k-means within row groups) to ARB-LLM (ablation) | ARB-LLM_RC · Llama-2-7B WikiText2 | perplexity: 16.44 -> 8.77 | Improved |
| add SimS row grouping (shared bitmap) without GBS (ablation) | ARB-LLM_RC · Llama-2-7B WikiText2 | perplexity: 16.44 -> 21.67 | Hurt |
| add GBS on top of SimS (ablation) | MemeBQ · Llama-2-7B WikiText2 | perplexity: 21.67 -> 16.73 | Improved |
| use greedy cosine-similarity grouping instead of merge-to-nearest (ablation) | SimS on BiLLM · Llama-2-7B WikiText2 | perplexity: 70,99 -> 31.91 | Improved |
| use greedy search instead of k-means for row grouping (ablation) | SimS on BiLLM · Llama-2-7B WikiText2 | perplexity: 37.68 -> 31.91 (cosine) | Improved |
| use cosine similarity instead of euclidean distance for row similarity (ablation) | SimS greedy search · Llama-2-7B WikiText2 | perplexity: 35.19 -> 31.91 | Improved |
| use cosine instead of euclidean for greedy grouping (ablation) | SimS greedy search · Llama-2-7B C4 | perplexity: 36.15 -> 37.24 | Mixed |
| increase group size from 1 to 2 (halve extra bits) (ablation) | MemeBQ · Llama-2-7B WikiText2 | perplexity: 8.77 -> 16.73 | Hurt |
| increase group size from 2 to 4 (0.25 extra bits) (ablation) | MemeBQ · Llama-2-7B WikiText2 | perplexity: 16.73 -> 33.81 | Hurt |
| disable bitmap (group size 0) (ablation) | MemeBQ · Llama-2-7B WikiText2 | perplexity: 8.77 -> 377.97 | Hurt |
| use MemeBQ instead of ARB-LLM_RC at the same extra bits | binary PTQ · Llama-3-8B 7 zero-shot QA average | average accuracy: 52.27 -> 59.54 | Improved |
| MemeBQ with half the extra bits versus ARB-LLM_RC | binary PTQ · Llama-2-7B WikiText2 | perplexity: 16.44 -> 16.73 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BiLLM — BiLLM: Pushing the Limit of Post-Training Quantization for LLMs (ICML 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- BiLLM: Pushing the Limit of Post-Training Quantization for LLMs (ICML 2024 · similarity 0.92)
- PB-LLM: Partially Binarized Large Language Models (ICLR 2024 · similarity 0.91)
- ARB-LLM: Alternating Refined Binarizations for Large Language Models (ICLR 2025 · similarity 0.91)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · similarity 0.91)
- Q-Palette: Fractional-Bit Quantizers Toward Optimal Bit Allocation for Efficient LLM Deployment (NeurIPS 2025 · similarity 0.91)
- ABQ-LLM: Arbitrary-Bit Quantized Inference Acceleration for Large Language Models (AAAI 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
