Deep Learning Explainer @DeepExplainer

Hi, I work on #NLProc and I make deep learning videos on Youtube, #deeplearning, #machinelearning youtube.com/channel/UC9aif… Joined July 2020

Tweets

518
Followers

314
Following

170
Likes

705

Deep Learning Explainer @DeepExplainer

2 years ago

come to our poster and say hi!

Tahmid Rahman @tahmedge

2 years ago

#EMNLP2024 paper alert: "Query-OPT: Optimizing Inference of Large Language Models via Multi-Query Instructions in Meeting Summarization" - to be presented on November 14 (Thu) 14:00-15:30 - Riverfront Hall! Let's explore how this saves time & cost.

1 2 8 843 1

0 0 1 117 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

When HuggingFace hits snooze, the entire LLM industry goes, "Guess it's time for an unexpected coffee break!"

0 0 1 104 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

the full paper "Tiny Titans: Can Smaller Large Language Models Punch Above Their Weight in the Real World for Meeting Summarization?" arxiv.org/pdf/2402.00841…

0 0 0 57 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

Fine-tuning a high-quality, smaller LLM can save you tons of money A fine-tuned FLAN-T5-780M can outperform zero-shot models up to 100 times its size, such as ChatGPT Check out our preprint "Tiny Titans: Can Smaller LLMs Punch Above Their Weight in the Real World?"

1 1 3 166 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@Yampeleg thanks for sharing and discussing our paper. Here is the link to the full paper arxiv.org/abs/2402.00841

0 0 2 20 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@ZachariahNKM @Yampeleg very interesting view point. it's worth investigating

0 0 0 27 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@jasonmeaux @Yampeleg that will be wild

0 0 1 15 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@Josh9817 @abacaj exciting to see the results

0 0 0 5 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@bert_on_spec @abacaj good finding!

0 0 2 6 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@graphsareprty @abacaj text summarization is one

0 0 1 8 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@CashCompounding @abacaj for text summarization at least

0 0 0 35 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@asusevski @abacaj It's not the perfect metric, but there are quite a few studies indicating that ROUGE scores align with human judgment relatively better than many other metrics.

0 0 0 44 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@mov_axbx @abacaj exactly. flan-t5 models are probably the most underrated deep learning models

0 0 1 45 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@thetaytay @abacaj @AiEleuther good to see more t5s :)

0 0 0 29 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@masta_g3 @abacaj There are few assumptions that that flan-t5 models are so good 1. Flan dataset (1000+ instruction tuning datasets) 2. encoder-decoder architecture I would like to see more big models instruction-tuned on Flan and more encoder-decoder LLMs

0 0 2 249 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@abacaj Thank you for sharing and discussing our paper. I am one of the authors. Here is the link to the paper arxiv.org/pdf/2402.00841…

0 0 1 14 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@n0riskn0r3ward @abacaj yes, they are from the same paper

0 0 2 10 0

View Details

Deep Learning Explainer @DeepExplainer

2 years ago

@rohanpaul_ai Thank you for the engagement with our paper. To address a frequently mentioned point, here's the exact excerpt from our study to present our findings in full clarity. This is the link to the paper arxiv.org/abs/2402.00841