Unverified50% confidenceBenchmarkExact time
在ESC-50数据集的声音分类实验中，训练前首位命中率约25%，在A100上训练约七分半钟后提升到约66%
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在演示数据集上训练15轮后，文本分类模型准确率达到99%，模型总参数11万（Attention 1.6万，FFN 3.3万，其余为Embedding）67% similarUnverified训练中保留集 FID 从 33.7 降到 27.0，FD-DINOv2 从 570 降到 218，基于检测器的物体准确率从 65% 提升到 90%66% similarUnverified小模型RL后训练中正确率从4%提升到45%（pass@1）65% similarVerified在CIFAR-10上训练40个epoch的复现实验中，plain-56的训练准确率为84.0%，plain-20的训练准确率为95.1%，尽管plain-56参数量（853,018）约为plain-20（269,722）的三倍64% similarUnverifiedCLM训练中仅预训练在困难负样本测试上得52%，加上困难负样本阶段跳升到69%；若从第一天就用困难负样本训练则峰值只有62%且很快过拟合；缺少重放会使困难负样本分数从69%跌回56%64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/989453API
curl https://kongchang.com/api/v1/knowledge/claims/989453MCP
get_claim(id=989453)