Unverified50% confidenceOpinionExact time
If 104B refers to activated parameters, this continues the sparsification approach: a larger parameter pool for capacity, and fewer activated parameters to control inference cost.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
Unverified104B activated corresponds to the number of parameters actually involved in computation per inference, which more directly determines latency and throughput levels.77% similarUnverified稀疏激活架构下推理阶段的FLOPs与激活参数规模线性相关而非与总参数规模相关,从而降低单次推理的计算资源消耗和延迟70% similarUnverified推理部署对单次计算规模要求低,但需要应对海量并发请求,更强调吞吐量和延迟的平衡69% similarUnverifiedThe bottleneck of on-device inference is often not peak compute but memory bandwidth; ternary quantization reduces the amount of data that needs to be read per inference by compressing parameters, which is key to achieving 120 tok/s.67% similarUnverifiedThe fundamental assumption of low-dimensional parameter mapping is that effective parameter updates tend to concentrate in a subspace far smaller than the full parameter space.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1092644API
curl https://kongchang.com/api/v1/knowledge/claims/1092644MCP
get_claim(id=1092644)