Barbara Gendron

PhD student in NLP @ LORIA, Nancy, France

[Barbara Gendron @BarbaraGen91392]

Still looking for best practices on #LoRA adapters config for efficient #LLM fine-tuning. Especially on Llama models. Any insight to share about it?
https://t.co/XQGfdusWzs

#HuggingFace #Transformers #NLProc

[Barbara Gendron @BarbaraGen91392]

@francoisfleuret Transformers 🤗

[Tom Dörr @tom_doerr]

3Blue1Brown's animation code is public

[Nat Friedman @natfriedman]

It is funny how consistently both Claude and ChatGPT output incorrect code for using the Claude and OpenAI APIs.

[Barbara Gendron @BarbaraGen91392]

@PTownAllStar93 @fchollet I think there is a major issue you are not considering here: making a model strongly specialized in several distinct domains is hard. Think about catastrophic forgetting. For now we can only hypothesize that AGI 1st training phase would mitigate it.

[Rohan Paul @rohanpaul_ai]

A Survey paper of Small Language Models

[BURKOV @burkov]

Python has become the most popular language on GitHub. I only wish for two more things: that browsers start supporting Python as a frontend language and that Excel fully supports it as a macro language. The world will be a better place.

[Xin Eric Wang @xwang_lk]

A fundamental question to ask for AI agent research: what is the major bottleneck that cannot be addressed by simply scaling up data and training next version of MLLMs?

[Michael Bronstein @mmbronstein]

Two Nobels for deep learning

[Omar Sanseviero @osanseviero]

BREAKING NEWS

The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Literature to the Attention Is All You Need authors.

Their work has made thousands cry, laugh, or rich and made GPUs go brrr

[Barbara Gendron @BarbaraGen91392]

@pmddomingos AI needs all of them and all of them need AI.

[Barbara Gendron @BarbaraGen91392]

@pmddomingos That might be a consequence of the second L meaning « language », just saying

[Barbara Gendron @BarbaraGen91392]

@mervenoyann 100% real

[Barbara Gendron @BarbaraGen91392]

@honutv_ @Inoxtag Pas peur de dire des conneries visiblement

[Barbara Gendron @BarbaraGen91392]

@SpectrGen Yes, that's right. The RAG principle is conceptually very simple, it can't make miracles :)

[Barbara Gendron @BarbaraGen91392]

@SpectrGen Thanks for your post, I was tired of reading all these posts saying RAG is wonderful! You're completely right + RAG setup is not suitable for multiple sources of info / coreferences / meta info about the data

[ᒍᑌᒪIᕮM🎶 @Wirewalkeuse]

Barbara Gendron est la première oratrice de cette session commune @jeptaln2024

[Barbara Gendron @BarbaraGen91392]

Ravie d’avoir pu prendre la parole en session commune de #jeptaln2024 hier pour parler de détection d’émotions en conversation #ERC 😁 Lire l’article : https://t.co/UnvPVgJ0u9

[Barbara Gendron @BarbaraGen91392]

@iamgingertrash @ylecun @alex_peys By saying « optimal answer », you assume that it exists and that we can evaluate the distance to optimality for any answer provided by the LLM. These are very strong assumptions to be fair…

[Barbara Gendron @BarbaraGen91392]

@mwaseema Yes it is.