ARB-LLM
LLM quantization
superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 1 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites ARB-LLM as a baseline.
“Existing binarization methods predominantly focus on weight-only designs while overlooking the quantization characteristics of activations, leading to suboptimal performance when activations are quantized to low bit-widths.”
— BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
Beaten on benchmarks
Head-to-head results where a newer method reports beating ARB-LLM. Values are copied from the source paper's tables — verify against the cited paper.
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA2-7B, 16-bit activation]
52.46 vs 36.74
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA2-13B, 16-bit activation]
62.38 vs 50.33
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA2-70B, 16-bit activation]
73.78 vs 63.69
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA3-8B, 16-bit activation]
48.81 vs 38.24
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [Qwen3-8B, 16-bit activation]
52.11 vs 42.51
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [Qwen3-14B, 16-bit activation]
62.65 vs 53.13
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [Qwen3-32B, 16-bit activation]
68.29 vs 65.67
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA2-7B, 6-bit activation]
42.05 vs 34.88
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA2-13B, 6-bit activation]
60.07 vs 42.29
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA2-70B, 6-bit activation]
72.38 vs 36.80
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [LLaMA3-8B, 6-bit activation]
45.79 vs 35.45
- BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
BWLA beats ARB-LLM · Avg. [Qwen3-8B, 6-bit activation]
50.46 vs 26.98
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.
- STaR-QuantSTaR-Quant: State-Time Consistent Post-Training Quantization for Diffusion Large Language ModelsJun 3, 2026
- May 26, 2026
- May 1, 2026
- Bit-by-BitBit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMsApr 9, 2026
- Benford-QuantBenford's Law as a Distributional Prior for Post-Training Quantization of Large Language ModelsJan 29, 2026
- HestiaHESTIA: A Hessian-Guided Differentiable Quantization-Aware Training Framework for Extremely Low-Bit LLMsJan 28, 2026
- Layer-Wise High-Impact Parameter Ratio OptimizationLayer-Wise High-Impact Parameter Ratio Optimization in Post-Training Quantization for Large Language ModelsNov 21, 2025
- Sep 28, 2025