const currentPage = dv.current().file;
const dailyPages = dv.pages('"0-Daily"').sort(k=>k.file.name, "asc");
const currentPageName = currentPage.name;
const index = dailyPages.findIndex((e) => {return e.file.name === currentPageName});
if (index < 1) {
dv.table(["File", "Created", "Size"],[]);
} else {
const lastIndex = index - 1;
const lastPage = dailyPages[lastIndex].file;
const allPages = dv.pages().values;
const searchPages = [];
const lastTime = dv.parse(lastPage.name);
const currentTime = dv.parse(currentPage.name);
debugger;
for (let page of allPages) {
const pageFile = page.file;
if (pageFile.cday > lastTime && pageFile.cday <= currentTime) {
searchPages.push(pageFile);
}
}
dv.table(["File", "Created", "Size"], searchPages.sort((a, b) => a.ctime > b.ctime ? 1 : -1).map(b => [b.link, b.ctime, b.size]));
}
探索上下文更多的模型
在网上冲浪的时候，发现 Mosaic-ML , 官方介绍 里讲
- Licensed for commercial use(unlike LLaMA).
- Trained on a large amount of data(1T tokens like LLaMA vs. 300B for Pythia, 300B for OpenLLaMA, and 800B for StableLM).
- Prepared to handle extremely long inputsthanks toALiBi(we trained on up to 65k inputs and can handle up to 84k vs. 2k-4k for other open source models).
- Optimized for fast training and inference(viaFlashAttentionandFasterTransformer)
- Equipped with highly efficient open-source training code.
能够支持最多 65k 的输入。比现在的 4k 多得多得多。
在 claude 目前没有 API 的情况下，尝试一下。
试用
MPT-7B
没有反应
chatgpt
脚本
https://colab.research.google.com/drive/1nDVSUEoW5lsmjiVCpHokP15ozc4Jqj-i?usp=share_link
响应
简单试验了一下， 效果不好。这种简单的总结都总结不出来。
总结
还是商业开源的好， GPT-3.5-turbo 效果应该是比开源的好得多的多。
后面等等 Claude 确认下吧。