Unverified50% confidenceFactExact time
Ternary quantization restricts neural network weights to three discrete values (typically -1, 0, and +1), requiring theoretically only about 1.58 bits of storage per weight.
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedQ3 quantization means each parameter is stored using approximately 3 bits73% similarUnverifiedTernary representation theoretically requires only about 1.58 bits per parameter, which can yield nearly a 10x storage compression compared to FP16.71% similarUnverified1.58-bit refers to each weight occupying an average of log₂(3) ≈ 1.58 bits, because ternary weights (−1, 0, +1) have 3 possible states.69% similarUnverifiedWhen a large portion of the model is offloaded to the CPU, any quantization bit-width that cannot be divided evenly by 2 incurs a speed penalty, making 3-bit actually slower than 4-bit.67% similarUnverified微软研究院2023年提出的BitNet b1.58架构将所有权重限定为{-1, 0, +1}三个状态，理论上可将模型大小压缩至原来的1/16以下67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/933016API
curl https://kongchang.com/api/v1/knowledge/claims/933016MCP
get_claim(id=933016)