Unverified70% confidenceFactExact time
NVFP4 introduces a per-group scaling factor mechanism, typically sharing one FP8 scaling factor per group of 128 weights
2
Sources
70%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Related Claims
UnverifiedHopper supports FP8, but its FP8 scaling is at tensor granularity, requiring a global scaling factor to be pre-computed before each matrix multiplication.61% similarUnverified模型量化是将模型权重从FP16压缩至INT4/INT8以节省显存59% similarUnverifiedThe hierarchical Gaussian process introduces a group-level structure on top of the GP: each post-stratification cell has its own group effect, while the group effects are related to one another through a shared kernel structure.58% similarUnverifiedNF4 is an information-theoretically optimal 4-bit quantization data type, based on the assumption that pre-trained neural network weights approximately follow a zero-mean normal distribution, dividing the normal distribution CDF into 16 equal intervals as quantization levels.57% similarUnverifiedQLoRA论文提出的NF4针对预训练权重服从正态分布的特性设计了最优量化格点57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/185207API
curl https://kongchang.com/api/v1/knowledge/claims/185207MCP
get_claim(id=185207)