Unverified50% confidenceBenchmarkExact time
在LongMemEval基准测试中，5个chunk时BM25准确率从78.8%提升到85.2%
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified本地运行的Gemma 26B在453次评分预测中准确率仅为49.2%，其配对的永远说涨基准准确率为53.4%74% similarUnverifiedEfficientNet 从 B0 到 B7，模型参数从 5.3M 增长到 66M，Top-1 精度从 77.1% 提升到 84.3%73% similarVerifiedCoT提示将PaLM 540B在GSM8K数学基准测试中的准确率从17.9%提升至58.1%71% similarUnverifiedAS5048 数据手册中标注的实际精度（考虑积分非线性误差 INL）通常在 ±0.5 度左右69% similarUnverifiedGLM-5.3的智能体编程准确率随投入词元增加而提升：低投入约24.7，高投入31.5，最大投入34.769% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/920796API
curl https://kongchang.com/api/v1/knowledge/claims/920796MCP
get_claim(id=920796)