Unverified50% confidenceFactExact time
Bonsai uses ternary quantization, compressing each weight to one of three values—-1, 0, or +1—with a shared scaling factor per 128 weights, requiring approximately 1.7 bits per weight, roughly 9 times smaller than 16-bit full precision.
1
Sources
50%
Confidence
Long-term
Relevance
10/4/2026
First Seen
Sources
Related Entities
Related Claims
Unverified1-bit Bonsai achieves an intelligence density score of 0.53/GB, which is more than 10 times that of the full-precision model and approximately 2.7 times that of the best low-bit alternatives73% similarUnverifiedTernary 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.72% similarUnverifiedTernary representation theoretically requires only about 1.58 bits per parameter, which can yield nearly a 10x storage compression compared to FP16.72% similarUnverifiedTaking a layer with a 3×3 convolution kernel and 256 input and output channels as an example, a standard convolution requires approximately 590,000 parameters, while depthwise separable convolution requires only approximately 67,000 parameters, a reduction of roughly 8–9 times.68% similarUnverified1-bit量化（BitNet）将每个权重压缩为单个比特，仅表示+1或-165% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/973446API
curl https://kongchang.com/api/v1/knowledge/claims/973446MCP
get_claim(id=973446)