Unverified50% confidenceFactExact time
Work such as Chinchilla points out that under a given compute budget, parameter count and training token count should grow roughly in proportion to achieve optimal efficiency.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedIn 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 1:2069% similarUnverified推理成本会随用户量线性乃至超线性增长，而训练是一次性或周期性成本69% similarUnverified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64% 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.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1083391API
curl https://kongchang.com/api/v1/knowledge/claims/1083391MCP
get_claim(id=1083391)