Verified90% confidenceFactExact time
In 2022, DeepMind published the Chinchilla paper, which corrected the optimal training ratio and indicated that, under a given compute budget, the number of model parameters and the number of training tokens should maintain an approximate ratio of 1:20
12
Sources
90%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Java程序员转型AI应用开发：航空智能客服实战指南
bilibili巴卜大模型7/6/2026
Related Claims
Unverified训练FLOPs可通过Chinchilla缩放定律估算，计算量约等于6乘以参数量乘以训练token数72% similarUnverifiedPost-training costs only 1–5% of the cost of training a large model from scratch, yet can yield a 20–50% performance improvement on target tasks.67% similarUnverifiedAnthropic's peak compute consumption during model pre-training is under 200 megawatts, with a duration of approximately two months.67% similarUnverified主流大模型的训练数据通常有6-18个月的滞后期67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/149732API
curl https://kongchang.com/api/v1/knowledge/claims/149732MCP
get_claim(id=149732)