繼上次把Forge裝起來後，我還是覺得生圖的方式太沒效率了，所以轉向尋求其他工具。這次嘗試的是Dify跟ComfyUI的結合。
(more...)
自從「布丁布丁吃什麼？」改用Felo AI作為站內的搜尋引擎之後已經一段時間了。這對部落格的使用帶來什麼影響呢？讓我們一起來看看吧。
Since "Pulipuli's Blog" switched to using Felo AI as its internal search engine, some time has passed. What impact has this had on the blog's usage? Let's take a look.
(more...)
儘管大型語言模型的盛行讓大家瞭解到AI分析文字的能力，但直接以命名實體識別和情緒分析來處理文本的做法，在具有特定目標的研究報告裡仍扮演著重要的角色。這篇就讓從我們從學習Python的使用，一步一步踏入命名實體識別和情緒分析的領域吧。
Although the prevalence of large language models has made people aware of AI's ability to analyze text, direct processing of text using Named Entity Recognition (NER) and Sentiment Analysis (SA) still plays a vital role in research reports with specific objectives. This article will guide us through the process of learning Python, step by step, to delve into the realm of Named Entity Recognition and sentiment analysis.
Short URL: https://l.pulipuli.info/24/nsysu
(more...)
由於這次RAGFlow看起來又無法順利完成任務了，我還是來記錄一下目前的狀況吧。
(more...)
我在「自行架設大型語言模式應用程式：Dify」這篇講到我用SerpAPI作為Dify的搜尋引擎，但除了使用別人提供的API之外，我們也可以用SearXNG自行架設客製化的搜尋引擎，並將它跟Dify結合一起使用。
In the article "Self-Hosting a Large Language Model Application: Dify," I mentioned using SerpAPI as the search engine for Dify. However, besides using third-party APIs, we can also utilize SearXNG to set up a customized search engine and integrate it with Dify.
(more...)
今天把電燈鎖緊，它亮了起來。但過一段時間，它又鬆脫了。
(more...)
為了讓大家瞭解如何使用提示詞來進行AI繪圖，本次教學以Kera AI為繪圖平台，一步一步帶領大家從簡單的提示詞、提示詞的變化，到使用圖片來引導繪製圖片。最後則是介紹更進階的控制圖片技巧，希望成為大家踏入AI繪圖領域的敲門磚。
To help everyone understand how to use prompts for AI art generation, this tutorial uses Kera AI as the drawing platform and will guide you step-by-step from simple prompts, variations of prompts, to using images as references. Finally, we will introduce more advanced image control techniques, hoping to serve as a stepping stone for everyone entering the field of AI art.
Short URL: https://l.pulipuli.info/24/hsc
(more...)
由於放在弱電箱裡面的無線基地臺一直當機，這次索性購買了MikroTik路由器RB450Gx4來用用看。這款號稱最適合放在弱電箱裡面的路由器，究竟會是如何呢？
(more...)
雖然大家都知道RAG可以將檢索結果交給大型語言模型回答，不過到底交給大型語言模型的是那些檢索結果？這些檢索結果又對應到那些文件？LLM開發平臺「畢昇」在檢索功能漂亮地解決了上述的問題，應可成為RAG應用中值得參考的標杆。
While everyone knows that RAG can submit retrieval results to large language models (LLMs), what exactly are those retrieval results submitted to LLMs? And which documents do these retrieval results correspond to? The LLM development platform, BISHENG, elegantly addresses these questions in its retrieval function and can serve as a valuable benchmark for RAG applications.
(more...)
我的課程網頁從Google Doc、Google Doc網頁、Google Doc Publisher，到現在終於進入到Google Sites了！這篇就來講講這段期間我設置課程網頁方案的演進吧。
(more...)
大型語言模型(Large Language Model)成為AI浪潮之後下一個新的寵兒，它彷彿真人般的對談和創造力的發想對研究和教育上帶來了無數啟發。但是大型語言模型在要求精確的工業領域裡面，究竟可以扮演什麼角色呢？本次演講先講述工業5.0發展中對於大型語言模型的需求，再來講述工業領域應用大型語言模型的實例，最後介紹大型語言模型和檢索生成增強的相關技術作為結尾。如果你也想在產業應用大型語言模型的話，不妨先看看這份投影片，瞭解一下現況吧。
Large Language Models (LLMs) have become the next big thing in the wake of the AI wave, offering human-like conversation and creative brainstorming that have inspired countless research and educational endeavors. But what role can LLMs play in demanding industrial fields that require precision? This presentation will first discuss the need for LLMs in the development of Industry 5.0, followed by examples of LLM applications in industrial settings. Finally, it will conclude with an introduction to related technologies like Retrieval-Augmented Generation. If you are also interested in applying LLMs in industry, take a look at this presentation to understand the current landscape.
Short URL: https://l.pulipuli.info/24/nkust
(more...)
這份投影片對檢索生成增強(Retrieval-Augmented Generation, RAG)的觀念作一個容易理解的介紹，也是「資訊檢索的AI革新：從資訊檢索到檢索增強生成」這篇的簡化版本。一般說到AI大家都會想到創造力、彷彿真人的表現，但RAG本質上更接近資訊檢索的問題。把它當作資料庫就很容易理解RAG的用途了。
This presentation provides an accessible introduction to the concept of Retrieval Augmented Generation (RAG), and is a simplified version of "The AI Revolution in Information Retrieval: From Information Retrieval to Retrieval-Augmented Generation". When people talk about AI, they often think of creativity and human-like performance, but RAG is essentially closer to the problems of information retrieval. Thinking of it as a database makes it easier to understand the purpose of RAG.
Fixed Short URL: https://l.pulipuli.info/24/nccu/rag
(more...)
很多人以為RAG可以取代搜尋引擎，但其實很多人要的功能只有「能用自然語言檢索」而已。
(more...)
Google NotebookLM已經成為使用RAG的最基本入門產品了。它大幅拉高了RAG的競爭門檻。沒有做的比Google NotebookLM好的話，都沒資格出來賣錢了。這篇是我在介紹Google NotebookLM的投影片，供大家參考。
Google NotebookLM has become the most basic entry-level product for using RAG. It has significantly raised the competitive bar for RAG. If it's not better than Google NotebookLM, it's not worth selling. These are my slides introducing Google NotebookLM for your reference.
(more...)
最近我因為研究需求蒐集了公視新聞網地方新聞的一些內容，並把資料整理表格資料集，提供給有需要的人使用。
Recently, for my research, I collected local news content from the PTS News website and organized the data into a tabular dataset, making it available for anyone who needs it.
(more...)
很多人都說跑大型語言模型需要很高級的GPU，其實相對於門檻較高的大型語言模型，小型語言模型也一直在如火如荼地發展。最近我嘗試用12核CPU跟32GB的RAM來跑Gemma2:2B，意外地很順利呢。
Many people say that running large language models requires high-end GPUs. However, relative to the higher barrier to entry of large language models, small language models have also been developing rapidly. Recently, I experimented with running Gemma2:2B using a 12-core CPU and 32GB of RAM, and it went surprisingly smoothly.
(more...)
我最近試用了SeaSalt公司底下的SeaMeet跟SeaChat，前者可以自動記錄並摘要 Google Meet會議內容，後者則是簡單建立聊天機器人。這篇就是我報告投影片的彙整，供大家參考。
(more...)
我開發了一份Colab筆記本「內部一致性信度分析：Cronbach's Alpha計算器」，方便大家分析量表的內部一致信信度Cronbach's Alpha。這份筆記本能夠分析整份量表的Cronbach's Alpha，也可以各別計算量表中各個構面(subscaler)的Alpha係數。同時計算器還會用逐一移除題目的方式，藉由指出移除後能夠提高Cronbach's Alpha的題目，讓研究者仔細檢視量表的設計是否有改善空間。
I developed a Colab notebook titled "Internal Consistency Reliability Analysis: Cronbach's Alpha Calculator" to facilitate the analysis of a scale's internal consistency reliability using Cronbach's alpha. This notebook can analyze the Cronbach's alpha for the entire scale, as well as calculate the alpha coefficient for individual subscales. The calculator also performs item-level analysis, identifying items whose removal would increase Cronbach's alpha, allowing researchers to carefully examine and potentially improve the scale's design.
(more...)
Nextcloud的AI應用程式不能處理中文，所以我自己用Langflow整合到Nextcloud，讓大型語言模型能夠根據我在雲端硬碟裡面的內容來回答問題。這篇就講一下大致上的做法。
(more...)
使用Colab來進行資料分析時，必須要經過程式執行的動作，才能取得執行後的分析結果。以下就讓我來看看怎麽在Colab裡面執行程式。
When performing data analysis using Colab, you must execute the code to obtain the analysis results. Let's take a look at how to execute codes within Colab.
(more...)
由於之前硬碟毀損，導致我用來做AI繪圖的Stable Diffusion環境全部消失。這次乾脆全部重來，用Stable Diffusion WebUI Forge重建整個繪圖環境吧。
(more...)
在Colab分析資料是現在很常用的技巧。但要將檔案上傳到Colab中，你需要進行以下的操作。
Analyzing data in Colab is a common technique nowadays. However, to upload files to Colab, you need to perform the following steps.
(more...)
在陸陸續續地升級了Dify的各個元件後，赫然發現，目前作為RAG的知識庫，仍有許多不足之處。到底RAG知識庫應該具備什麼功能？這篇就來聊聊我的想法。
(more...)
Dify的知識庫能夠取自網路資料，再搭配我們自架的Coolcrawl來取代公開服務Firecrawl，就能夠抓取內部區域網路裡面的網路資料。接下來就讓我們來看看這要怎麼實作吧。
Dify's knowledge base can draw from online data, and by combining it with our self-hosted Coolcrawl (instead of the public service Firecrawl), it can also crawl data within the intranet. Let's take a look at how to implement this.
(more...)
雖然威技的WIFI除溼機的確能夠透過Tuya物聯網跟Home Assistant連接，但它似乎會在網路中斷的時候，直接關閉WIFI功能的樣子。原本以為除溼機都能按照自動化規則自己好好運作，怎麽不知不覺又離線了呢？
(more...)
唸資圖系或圖資系，未來可以做什麼工作？雖然大家的說法眾說紛紜，不過「圖書館員」仍然是此領域人才訓練的主要目標。在台灣要成為圖書館員的各種途徑中，透過國家考試成為具有「圖書資訊管理」專業的公務人員，又是許多人理想中的圖書館員職位。那要怎麼透過公職考試考上館員？要讀什麼書？要怎麼準備？我們就用這份講座來跟大家說明吧！
What kind of jobs can you get with a degree in Library and Information Science? While there are many different opinions, "librarian" remains the primary career goal for individuals trained in this field. In Taiwan, among the various paths to becoming a librarian, securing a civil servant position specializing in "Library and Information Management" through national examinations is the ideal librarian role for many. So, how can one pass the civil service exam to become a librarian? What should one study? How should one prepare? Let's address these questions in this lecture.
Fixed Short URL: https://l.pulipuli.info/24/dils/moe
(more...)
「布丁布丁吃什麼？」的總瀏覽量突破1500萬了！趁這個新的一年，一起讓我們來回顧看看2024年「布丁布丁吃什麼？」有哪些熱門話題吧！
(more...)
用關鍵字檢索站內資訊時，你總是看到大量分散的網頁，不知道怎麽整合最需要的資料嗎？這篇教你用Felo Search來建立站內的「問答機器人」！
Are you tired of keyword searches returning tons of scattered web pages, making it difficult to consolidate the information you need? This post will teach you how to use Felo Search to create an internal "Q&A chatbot"!
(more...)
Comments