Unverified50% confidenceFactExact time
Davinci has approximately 175 billion parameters, while Curie, Babbage, and Ada progressively reduce parameter scale, corresponding respectively to scenarios such as classification, semantic search, and text embedding.
1
Sources
50%
Confidence
Long-term
Relevance
9/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe model places n-gram lookup tables at layers 1 and 14, with 24 hashes each indexing approximately 16 million rows, totaling 196 billion parameters, exceeding one-third of the model's weights.63% similarUnverifiedMoE架构中，一个标称800亿参数的模型每次推理时可能只激活约200亿参数，VRAM占用大幅降低61% similarUnverifiedFor a configuration of 1.6 trillion total parameters and 49 billion activated parameters, only about 3% of parameters participate in computation during actual inference.60% similarUnverifiedDistilBERT将参数量压缩至6600万（原版1.1亿），推理速度提升60%，精度损失不足3%57% similarUnverifiedTongyi Qianwen 3.8 has approximately 2 trillion parameters, and Kimi has approximately 2 trillion parameters.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/955833API
curl https://kongchang.com/api/v1/knowledge/claims/955833MCP
get_claim(id=955833)