Unverified50% confidenceFactExact time
The conclusion of the Chinchilla scaling law: under a fixed compute budget, model parameter count and training token count should scale roughly in a 1:20 ratio in tandem, and early models were generally data-hungry.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2022年DeepMind发布的Chinchilla论文修正了原始Scaling Laws，指出应在给定计算预算下同比例扩展参数量和训练数据量71% similarUnverifiedIf samples exceeding 32K tokens rarely appear in a model's pre-training corpus, the model's performance in that range tends to be unstable even when the context window is extended through positional encoding.68% similarUnverifiedWith a fixed learning_rate, when batch_size scales linearly with the number of environments and num_epoch remains unchanged, the total displacement pushing the policy toward a new distribution in a single iteration may increase, manifesting as violent oscillation or even collapse of rewards in the early stages of training.68% similarUnverifiedBetween the two runs the team adjusted the generation budget (the first run occasionally hit the token limit before answering), so part of the fluctuation may come from the run framework rather than the model's judgment itself.68% similarUnverified插值阈值在数学上对应模型参数数量等于训练样本数量的临界点，此时线性方程组恰好有唯一解67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1051002API
curl https://kongchang.com/api/v1/knowledge/claims/1051002MCP
get_claim(id=1051002)