Corentin Dancette (@cdancette) 's Twitter Profile
Corentin Dancette

@cdancette

Machine Learning Researcher @Raidium_med. PhD in Deep Learning from Sorbonne Université. Graduated from MS. at @GeorgiaTech and @CentraleSupelec.

ID: 22513981

linkhttps://cdancette.fr calendar_today02-03-2009 18:15:58

295 Tweet

303 Takipçi

229 Takip Edilen

Corentin Dancette (@cdancette) 's Twitter Profile Photo

The first open "Audio LLM", by Kyutai, is available for demo! Also check out the announcement video for examples, it's really impressive, congrats to the team! youtube.com/live/hm2IJSKcY…

Tri Dao (@tri_dao) 's Twitter Profile Photo

FlashAttention is widely used to accelerate Transformers, already making attention 4-8x faster, but has yet to take advantage of modern GPUs. We’re releasing FlashAttention-3: 1.5-2x faster on FP16, up to 740 TFLOPS on H100 (75% util), and FP8 gets close to 1.2 PFLOPS! 1/

FlashAttention is widely used to accelerate Transformers, already making attention 4-8x faster, but has yet to take advantage of modern GPUs. We’re releasing FlashAttention-3: 1.5-2x faster on FP16, up to 740 TFLOPS on H100 (75% util), and FP8 gets close to 1.2 PFLOPS!
1/
AI at Meta (@aiatmeta) 's Twitter Profile Photo

Introducing Meta Segment Anything Model 2 (SAM 2) — the first unified model for real-time, promptable object segmentation in images & videos. SAM 2 is available today under Apache 2.0 so that anyone can use it to build their own experiences Details ➡️ go.fb.me/p749s5

David Picard (@david_picard) 's Twitter Profile Photo

Avec le prix Nobel de physique pour l'apprentissage et les réseaux de neurones, il est clair qu'on est face à un changement majeur (type électricité, nucléaire, télécom). Il nous faut une politique nationale plus ambitieuse! Jean Zay est déjà trop petit! Genci Ministère Enseignement supérieur et Recherche

Mustafa Shukor (@mustafashukor1) 's Twitter Profile Photo

We release AIMv2, a major step in scaling vision encoders. Properly scaling vision encoders has been challenging and lagging, compared to LLMs. The main bottleneck is training and evaluating on single image modality, (1/n)

We release AIMv2, a major step in scaling vision encoders. 
Properly scaling vision encoders has been challenging and lagging, compared to LLMs. The main bottleneck is training and evaluating on single image modality, (1/n)
Theo - t3.gg (@theo) 's Twitter Profile Photo

Hot take: I rarely if ever do "git add *" or "git add ." "git add -p" is super underrated. You basically do a mini code-review before making the commit. Essential part of my workflow

Jeff Dean (@jeffdean) 's Twitter Profile Photo

Want to check out the source for the "AlexNet" paper? Google has made the code from Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton's seminal "ImageNet Classification with Deep Convolutional Neural Networks" paper public, in partnership with the Computer History Museum. As I

Want to check out the source for the "AlexNet" paper?  Google has made the code from Alex Krizhevsky, <a href="/ilyasut/">Ilya Sutskever</a>, and <a href="/geoffreyhinton/">Geoffrey Hinton</a>'s  seminal "ImageNet Classification with Deep Convolutional
Neural Networks" paper public, in partnership with the Computer History Museum.

As I
Mustafa Shukor (@mustafashukor1) 's Twitter Profile Photo

We release a large scale study to answer the following: - Is late fusion inherently better than early fusion for multimodal models? - How do native multimodal models scale compared to LLMs. - How sparsity (MoEs) can play a detrimental role in handling heterogeneous modalities? 🧵

We release a large scale study to answer the following:
- Is late fusion inherently better than early fusion for multimodal models?
- How do native multimodal models scale compared to LLMs.
- How sparsity (MoEs) can play a detrimental role in handling heterogeneous modalities? 🧵
Mustafa Shukor (@mustafashukor1) 's Twitter Profile Photo

We are organizing a CVPR in Paris event the 6th of June. It will feature poster sessions for papers accepted at CVPR, and Keynotes from Alexei Efros, Diane Larlus and Alexandre Alahi. You can register here: cvprinparis.github.io/CVPR2025InPari…

We are organizing a CVPR in Paris event the 6th of June. It will feature poster sessions for papers accepted at CVPR, and Keynotes from Alexei Efros, <a href="/dlarlus/">Diane Larlus</a> and <a href="/AlexAlahi/">Alexandre Alahi</a>.
You can register here: cvprinparis.github.io/CVPR2025InPari…
Mustafa Shukor (@mustafashukor1) 's Twitter Profile Photo

The Worldwide LeRobot hackathon is in 2 weeks, and we have been cooking something for you… Introducing SmolVLA, a Vision-Language-Action model with light-weight architecture, pretrained on community datasets, with an asynchronous inference stack, to control robots🧵

The Worldwide <a href="/LeRobotHF/">LeRobot</a>  hackathon is in 2 weeks, and we have been cooking something for you… 
Introducing SmolVLA, a Vision-Language-Action model with light-weight architecture, pretrained on community datasets, with an asynchronous inference stack, to control robots🧵