TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Current architectural best practices for LLM applications

141 pointsby burakemirabout 2 years ago

7 comments

triyambakamabout 2 years ago
This is super helpful. I&#x27;m building a document question-answering service over a custom data corpus (related to Saivism, a sect of Hinduism). So far the first pass has been to manually chunk the text (based on headings, chapters etc.) and then I&#x27;ve used OpenAI&#x27;s embedding service and storing the embeddings in Pinecone. All stiched together using LangChain. To ask a question, the question is again embedded, then searched against the vector store, then the related documents are provided as context to the LLM along with the question.<p>So far it was really easy to set up the prototype, but the results weren&#x27;t as great as I had hoped, so I&#x27;m excited to see how I could improve it.<p>Edit: wow, I didn&#x27;t see this before. LangChain implements one of the featured article&#x27;s suggestions (HyDE) - <a href="https:&#x2F;&#x2F;python.langchain.com&#x2F;en&#x2F;latest&#x2F;modules&#x2F;chains&#x2F;index_examples&#x2F;hyde.html" rel="nofollow">https:&#x2F;&#x2F;python.langchain.com&#x2F;en&#x2F;latest&#x2F;modules&#x2F;chains&#x2F;index_...</a>
评论 #35736227 未加载
评论 #35736291 未加载
vectoralabout 2 years ago
This is one of the areas of LLMs that I find most interesting. So far, I&#x27;ve found simple question-answering over vectorstores to be a lackluster experience. In particular, the more information you embed and stick into the vectorstore, the less useful the system becomes as you are less likely to get the information you&#x27;re looking for (especially if the users don&#x27;t understand their queries need to look like the docs the want to ask about.<p>I haven&#x27;t had a chance to try out hypothetical embedded docs yet, but I expect they only provide a marginal improvement (especially if QAing over proprietary data or information).<p>I&#x27;d love to see any other interesting, more up-to-date resources anyone has found on this topic. I found this recent paper interesting: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2304.11062" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2304.11062</a>
评论 #35735817 未加载
johntb86about 2 years ago
This document seems to have been written before the toolformer paper[0], which fine tunes the model to use tools (e.g search) to retrieve information.<p>[0]: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2302.04761" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2302.04761</a>
评论 #35736434 未加载
foxgroverabout 2 years ago
Few other helpful options recently added to Langchain:<p>1. Extraction for query filters - <a href="https:&#x2F;&#x2F;twitter.com&#x2F;hwchase17&#x2F;status&#x2F;1651617956881924096?s=46&amp;t=gkyxL9FAhSE-DiMAkwTkcg" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;hwchase17&#x2F;status&#x2F;1651617956881924096?s=4...</a><p>2. Contextual compression to eek more out of prompt stuffing - <a href="https:&#x2F;&#x2F;twitter.com&#x2F;hwchase17&#x2F;status&#x2F;1649428295467905025?s=46&amp;t=gkyxL9FAhSE-DiMAkwTkcg" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;hwchase17&#x2F;status&#x2F;1649428295467905025?s=4...</a><p>And then it’s been there’s existing great utility chains for map-reduce, with re-ranking, etc for more ways to apply LLM completions over large documents and&#x2F;or large sets of documents: 3. <a href="https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=f9_BWhCI4Zo">https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=f9_BWhCI4Zo</a>
gdiamosabout 2 years ago
We are going to need better retrieval methods as LLMs augment and generate more content on the internet.
评论 #35735867 未加载
KrugerDunningsabout 2 years ago
Sentence embeddings have been great for improving semantic search, but I am still struggling with finding relevant documents for numerical values. Questions like &quot;what people where born in 1992&quot; or &quot;people with at least 4 children&quot;. One thing I can do is pre-process the data by transforming the date of birth into boomers&#x2F;zoomers&#x2F;millenials and the like but this does not help on the question side if people don&#x27;t know what to ask
quickthrower2about 2 years ago
Is this architecture what autogpt uses?