Activity
Mon
Wed
Fri
Sun
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
Memberships
AI Developer Accelerator
Public • 3.8k • Free
5 contributions to AI Developer Accelerator
17d ago in
Trouble with Chroma DB retriever
I have some trouble with Chroma DB, specially in the minute 2:11:00 from the LangChain Master Class. I have done all the steps indicated, but when I try to retrieve the documents to do the query "How can I learn more of Langchain?", it retrieves documents that are non sense! I need some help to solve this, because on Stack Overflow some people said that they are just using Chroma library instead of the LangChain wrapper. The only thing I changed from the original code was the import, because it was deprecated and now it is recommended to use `from langchain_chroma import Chroma`. Does anyone has this same problem?
0
4
New comment 14d ago
0 likes • 16d
Chroma is not very friendly - as its openai-telemetry dependency does not lock down protobuf version. This can cause random application crashes - if you have protobuf version mismatches in your project dependencies. We (speaking with my professional head on) moved over to Faiss instead. YMMV. regarding receiving random "nonsense" responses - if you're using a RetrievalQAChain - set the value of q to be larger. Rule of thumb: less than 10 content pages - set q to 5, more than 10 content pages -> set q to 10 or above. RAG content generation (based on QA approaches) is not always accurate - but you should consider graphrag if you want improved accuracy (or hybridrag).
0 likes • 14d
Seems fine - don't expect the LLM to differentiate on roles - your app IMHO should manage what content is added to the retriever which the LLM is given. Content management/segmentation is usually something we do outside of the LLM step ( airflow /dagster etc). The LLM simply processes the content segment we choose to give it. Lots of ways to do this really but if I was setting up for a robust production and scaleable solution - I'd lean on these data workflow architectures for up front content management prior to LLM involvement.
22d ago in
Issue with Project Details Page on Vercel Deployment
Hi everyone! I'm running into an issue with lesson 13.7, "Vercel Blob Store Upload." I deployed the project to Vercel, and everything works fine locally. When running the app locally, selecting a project takes me to the project details page as expected. However, on the deployed Vercel version, it shows a "not found" page instead. Has anyone encountered this before? Any ideas on what I might be overlooking? Thanks in advance
1
4
New comment 16d ago
1 like • 16d
Download beyondcompare (I've no financial interest in this product btw). Then download Brandon's zip files of the current project state - that he leaves as a link at the bottom of some of the sections. Unzip and then use beyondcompare to compare diffs between your code, and Brandon's code - this will help you find minor discrepancies like typos etc. Note: that Brandon's code does not *always* mirror what is walked through in the videos - but its very close. [Brandon if you read this: audioprocessor/videoprocessor lesson - minor difference in the result dictionary being returned - your video has 3 keys in the dict, your final code download has 2 keys in the dict]. :)
16d ago in
401 errors in Railway app deployment logs
Remember to check your API_BASE_URL settings in your project variables *but also* go to your Vercel project settings and remember to turn off Vercel Authentication
0
0
Oct 7 in
Premium Course Support
Hi Fullstack AI Marketing Platform Course participant With 254 messages in our support thread, finding solutions has become challenging. To streamline this process, I propose we adopt a standardised format for issue reporting: Module: [Number and Name] Section: [Number and Name] Operating System: [Windows/Mac/Linux] Issue: [Brief description] Example: Module: 13. Asset Processing NextJS Section: 5. Manage Upload Header Operating System: Windows Issue: Sidebar doesn't stay at bottom of screen Benefits of this approach: 1. Easier searching for similar issues 2. Potential for creating a searchable FAQ database 3. Identifying recurring problems for @Brandon Hancock to address This format will help us all navigate and resolve issues more efficiently. What do you think about implementing this system?"
6
17
New comment Oct 20
0 likes • Oct 20
Hi everyone - anyone else get build errors with their first Vercel build/deployment ? Am happy to go and fix - but was expecting this to have built first time. [Failed to compile."Image" defined but not used; @typescript-eslint/no-unused-vars]. As stated - am happy to fix - just wondering if I have some module version misalignment or build configuration misalignment [following the lessons verbatim so far].
0 likes • Oct 20
*ignore - now fixed*. Added an extra rule to .eslintrc.json to turn off no-unused-vars check; also added a public folder and a public/.gitkeep file.
Sep 22 in
CrewAI Development Query
Hello all, am just curious to know if there's anyone here who is still developing and using crewai? It's one of the relatively easy ones that I could understand but the issues with it is their version isterations always inadvertently breaks some other functionality for some reason.
2
8
New comment Sep 27
0 likes • Sep 25
Pin your versions. Regressive breaks occur in many frameworks as they evolve.
1-5 of 5
@paul-mcgrath-6938
Computing professional. Keen to learn marketing AI domain/workflows, and other stacks/frameworks
Active 3h ago
Joined Sep 25, 2024
powered by