Unverified50% confidenceFactExact time
2.4万亿参数模型在BF16精度下考虑KV Cache和激活值后实际约需1.4-1.5TB显存
1
Sources
50%
Confidence
Long-term
Relevance
7/23/2026
First Seen
Sources
Related Claims
VerifiedQwen3-235B has a total parameter count of 235 billion, adopts a MoE architecture, and activates approximately 22B parameters during each forward inference pass.73% similarUnverified千问三235B MoE总参数量2350亿，但每次推理实际激活的参数约22B73% 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% similarVerifiedThe larger the number of model parameters, the greater the robustness to quantization, allowing 70B-level models to maintain performance close to FP16 after 4-bit quantization.69% similarUnverified万象模型提供了 1.3B 和 14B 参数量的不同规模版本69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/602483API
curl https://kongchang.com/api/v1/knowledge/claims/602483MCP
get_claim(id=602483)