Unverified50% confidenceFactExact time
The project adopts a sparse mixture-of-experts (MoE) architecture, in which only about 9B parameters are activated per token out of 74B total parameters, an activation ratio of about 12%.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
74B Model Squeezed into 24GB VRAM: Running Locally for Free on a Single 4090
bilibilijeffzhengye10/15/2026
Related Entities
Related Claims
UnverifiedMoE models activate only approximately 10 out of 512 experts per token, with an effective activated parameter count of approximately 20–30B.77% similarUnverifiedM2.5 adopts a Mixture of Experts (MoE) architecture, where only 10B parameters are actually involved in computation during each inference, while the total parameter count may far exceed 10B.74% similarVerifiedThe Mixture of Experts (MoE) architecture activates only approximately 10%–30% of total parameters during inference, significantly reducing FLOPs and GPU memory usage per inference73% similarVerifiedThe MoE architecture typically activates only 10%–20% of total parameters during each inference71% similarUnverifiedMoE的8选2配置意味着8个专家中每次只激活2个，实际计算量仅为密集模型的约四分之一70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1037458API
curl https://kongchang.com/api/v1/knowledge/claims/1037458MCP
get_claim(id=1037458)