Trying @perplexitycomet.bsky.social and testing the first "agentic" suggestion. You cant make this up.

denne tenkte jeg også på, så ble bare forvirra selv og mistet poenget ;)

heldigvis er en av fordelene med språkmodeller at de er veldig gode til å oversette til det språket man ønsker ;)

Preprint by @marijnvandermeer.bsky.social & @matthias-huss.bsky.social!

The MassBalanceMachine (MBM), a #MachineLearning model, predicts glacier mass balance at high resolution, even without in situ data. For Norwegian glaciers, MBM generalizes well & outperforms TI models in seasonal predictions.

i was thinking in terms of methodology. Im reading, but it takes some time 😅

i need sota LMs to handle comms w frens & fam so i can focus on my true passion of writing matplotlib code, reformatting latex tables and cleaning bibtex

😅

For someone new to this, how does these normalizing flows compare to the cnf and flow matching in lipman et al (2024)?

Weekend project: Implementing conditional normalizing flows in low dimensions from scratch

#mlsky

Men joda joda 😅

Summen kan jo bli det samme, bare mindre skatt på arbeid før det blir arv

Skattlegge rike folk som er døde, hvem kan være imot det? ;)

Skatt er en uting. Men alle er enige om at vi må finansiere staten på et eller annet nivå (med unntak av anarkistene)

not that anyone else is owning that verb

The paper will be presented on the #NLDL conference in Tromsø in January.

And I have successfully managed to confuse my bsky algo by mixing some personal paragliding and professional #LLM content here

The added benefit of being bayesian here is that tasks with less data will be more similar to the hierarchical mean parameter set Θ. And therefore learn from other tasks!

Effectively, each task adapter will be optimized both towards the training data, but also to be similar to the other adapters. And therefore the adapters will share knowledge between them!

It works by constructing a hierarchical LLM where each task adapter parameter set θ_d has a prior to a shared hierarchical mean parameter set Θ.

It outperforms both the case when you train a shared adapter for all tasks, and the case when you train one adapter per task indepedently on our dataset.

New finetuning method: Bayesian Hierarchical LoRa (BORA) (arxiv.org/abs/2407.15857)

If you have multiple and similar LLM tasks you want to finetune, you should share knowledge between the different adapters when training, and we show how using Bayesian Hierarchical modelling

Sunda problem preparing for the 2025 paragliding season: show temperature in altitude for all the weather stations in the area

You find an interesting plattform and immediately pipe it into your slow and legacy work-chat-platform?

Hello World from Python SDK