Unverified50% confidenceSolutionExact time
In speculative decoding, the draft model's parameter count is typically about 1/10 to 1/5 of the target model, and the candidate count k is typically 4-8.
1
Sources
50%
Confidence
Long-term
Relevance
9/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在MoE模型中每个token只激活少量Expert（通常2-4个，Top-K路由），实际计算量可能仅相当于6-7B的Dense模型72% similarUnverifiedMoE模型中'35B'指总参数量，每个token只激活少量Expert（通常2-4个），实际计算量可能仅相当于6-7B的Dense模型71% similarUnverified2.4万亿参数模型在BF16精度下考虑KV Cache和激活值后实际约需1.4-1.5TB显存71% similarUnverifiedLarge language models that can run on-device typically have parameter scales of 1B to 13B, far smaller than cloud-based models such as GPT-4 or Claude with hundreds of billions of parameters.68% similarUnverifiedLFM2.5-8B-A1B总参数量为8B，推理时仅激活1.5B参数67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/880306API
curl https://kongchang.com/api/v1/knowledge/claims/880306MCP
get_claim(id=880306)