Where can I find the list of MultiQC supported too...
# multiqc
e
Where can I find the list of MultiQC supported tools? https://docs.seqera.io/multiqc/modules doesn't seem to work
r
Do you not have this dropdown on the left side?
e
Damn I'm blind. Thanks
p
Yeah I need to rebuild this page to have a table with a filter at the top like on the old multiqc.info but haven't gotten around to it yet sorry
So just sidebar for now 👍🏻
e
That would help with searching the whole page, like the old version, aye!
p
Yeah it got put in the "after Summit" pile 😅
e
Extra work for all of us, you are not alone 😛
m
Indeed, a pity your tool now got fully seqeraized…I also already cursed the regressed current page (compared to the tidy beauty that was the previous Astro version)
💯 1
p
sorry, will get to it ASAP 🫡
m
But apart from the website, I think the new features you showcased at SBW yesterday are amazing! In case you can disclose: Was it require to additionally train a foundation model to be able to identify sequencing QC issues or does that still work with clever context management using MegaQC data?
p
Hi folks! Page is updated and back to its former glory: https://docs.seqera.io/multiqc/modules/
🙏 2
Sorry @Matthias Zepper I missed your follow-up message - no the demo I showed doesn't use any special foundation model. We run on top of the big providers (OpenAI, Anthropic) with various levels of RAG and fine-tuning etc.
The rest is all tooling and state management and stuff built on top of that to maintain context, history and a smooth UI
I did a podcast with Sasha a month or two back where we sort of glossed over some of this, but it might not be in-depth enough to satisfy your curiosity 😅
m
Thanks for both! It gave me a rough idea and that is already better than nothing. During a recent DDLS Hackathon, I played around a bit with LLMs and RAG, predominantly to assist the project coordinators, but generally folks at NGI are super sceptical about it. If at all, it must be an entirely self-hosted application, so no external providers.
p
Then the results will be much shittier. The big providers are far, far ahead of anything you can self-host.
But don't confuse public ChatGPT with openAI contracts with API use. The two are quite different in their policies and attitude to data privacy
Microsoft have some ways to get large LLMs with data residency by running on azure I think 🤔
m
I am aware and I think Amazon Bedrock is actually providing quite a lot building blocks that we could use. But it is a mixture of “not enough time” and not really a clear vision what we could or would want to achieve with it.
👍🏻 1