QTIP (LLM quantization): superseded — cited as a baseline and beaten by newer methods. 1 paper(s) critique it, 2 beat it on benchmarks — #25 of 80 most-superseded. Sub-problem: cluster led by RTN. Newer alternatives in the same sub-problem include STaR-Quant, Timestep-Aware SVDQuant-GPTQ, BWLA, Bit-by-Bit, Benford-Quant.

LLM quantization

QTIP

QTIP: Quantization with Trellises and Incoherence Processing

Superseded baseline#25 of 80 most-superseded · first seen Jun 17, 2024

Superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 2 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites QTIP as a baseline.

Although QTIP shows significant improvement over TCQ, it still suffers from high computational complexity.

— CCQ: Convolutional Code for Extreme Low-bit Quantization in LLMs

Beaten on benchmarks

Head-to-head results where a newer method reports beating QTIP. Values are copied from the source paper's tables — verify against the cited paper.

GLVQ-32D beats QTIP
3.36 vs 3.78
Perplexity · [2-bit, Llama 2-70B]
Learning Grouped Lattice Vector Quantizers for Low-Bit LLM Compression
ICQuant^SK-8.25% beats QTIP
8.25 vs 8.96
C4 · [Llama2-7B, ctx. 4096, 2.4 bits]
ICQuant: Index Coding enables Low-bit LLM Quantization
GLVQ-8D beats QTIP
40.0 vs 39.2
ARC-Challenge · [2-bit quantization, Llama 2-13B]
Learning Grouped Lattice Vector Quantizers for Low-Bit LLM Compression
ICQuant^SK-5% beats QTIP
6.26 vs 6.28
C4 · [Llama2-13B, ctx. 4096, 3.3 bits]
ICQuant: Index Coding enables Low-bit LLM Quantization

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.