Unverified50% confidenceFactExact time
Floating-point addition is not associative, and the bf16 format has only a 7-bit mantissa, so rounding errors accumulate across multi-layer multiply-add operations and may cause the argmax result to flip.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the validation of 42 offset and shift combinations, the maximum deviation of RoPE relativity was 1e-15, which is at the level of floating-point error.67% 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66% similarUnverifiedBF16 uses 1 sign bit, 8 exponent bits, and 7 mantissa bits, sacrificing precision while retaining the same dynamic range as FP32.65% similarUnverifiedThe 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.65% similarUnverifiedFor 32-bit integers exceeding 2²⁴, the mantissa precision of single-precision floating-point is insufficient to represent the value exactly, causing the exponent field to be underestimated.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1060910API
curl https://kongchang.com/api/v1/knowledge/claims/1060910MCP
get_claim(id=1060910)