Unverified50% confidenceFactExact time
The degree-3 and degree-4 polynomial coefficients in the paper are obtained by minimizing the maximum error (minimax approximation) in the BF16/FP16 target format.
1
Sources
50%
Confidence
Long-term
Relevance
10/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe FP4 format has only 4 bits (1 sign bit, 2 exponent bits, 1 mantissa bit) and can precisely represent only approximately 16 discrete values.72% similarUnverifiedINT4量化将原本由65536个不同值（FP16）表示的权重压缩到仅16个离散级别71% similarUnverifiedThe authors tested degree-3 and degree-4 BF16 polynomial versions to balance accuracy and speed.70% similarUnverified在4-bit量化下，纯参数理论最小占用为：7B约3.26GiB，13B约6.05GiB，32B约14.90GiB，70B约32.60GiB69% similarUnverifiedGPTQ (2022) achieved the first usable performance of large models at INT4 by quantizing layer by layer and minimizing quantization error using Hessian information.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/971821API
curl https://kongchang.com/api/v1/knowledge/claims/971821MCP
get_claim(id=971821)