Clément Chadebec

Research Scientist @ Jasper Research | Ph.D in Applied Maths (Generative Models) @Inria. I also maintain python packages democratizing Deep Generative Models.

[Clément Chadebec @CChadebec]

After 8 months of long coding nights ☕️ we finally officially release Pythae 🥳, a python library unifying generative autoencoder implementations including vaegan🥗, vqvae or RAEs.

🖥️ github repo: https://t.co/570oxztyn3
👉paper: https://t.co/Nh5BgRWtU7

[clem 🤗 @ClementDelangue]

Feels quite magical to be able to clone a 68 TB dataset to my private HF training bucket while I only have a 4TB local disk, all of that in less than a minute thanks to HF infra optimizations & xet dedup!

[Clément Chadebec @CChadebec]

@cyrildiagne Thank you Cyril!

[Cyril Diagne @cyrildiagne]

very nice dataset drop from the clipdrop team 🙌

[Julien Chaumond @julien_c]

We are starting to be quite bullish about getting in the data infrastructure business.

I just cloned 68 TB (while I only have a 4TB local disk) to my @huggingface training bucket in 1 minute 55 seconds, thanks to Xet deduplication and all our infra optimizations.

You can host your data processing pipelines on HF and leverage those insane optimizations 🔥

[Pedro Cuenca @pcuenq]

wow 🤯

68 TB, high-quality, high-resolution, Apache 2.

Thank you! 🙌🫡

[Clément Chadebec @CChadebec]

@JulienBlanchon @heyjasper @huggingface This may be an idea for future work indeed!

[Clément Chadebec @CChadebec]

📢 New @heyjasper release ! 📢

MONET 🌸 : An Apache2.0 deduped and recaptioned dataset of 105M samples unlocking reproducible text-to-image research.

Nano T2I 🖌️ : A codebase to train your own T2I model

🤗 @huggingface: https://t.co/x6gEhQIaFV
💻: https://t.co/K6VIU2wjtW

Very excited about this new release, pushing the boundaries of open and reproducible T2I research.

Congrats to the team!

Benjamin Aubin Gonzalo Quintana @onurxtasar @UlaLaParis @_jeev2 @dh7net @clipdropapp @heyjasperai

[Clément Chadebec @CChadebec]

@AmelieTabatta @heyjasperai Thank you for the support, means a lot! I hope it will be useful in your research

[Amélie Chatelain @AmelieTabatta]

Ah! this is insane, I've been complaining for weeks about the lack of open text-to-image data! Congrats on the release @heyjasperai

[Clément Chadebec @CChadebec]

@JulienBlanchon @heyjasper @huggingface Yes indeed! We hope this dataset can be useful for other tasks as well.

[Clément Chadebec @CChadebec]

@JulienBlanchon @heyjasper @huggingface Thank you!

[Wauplin @Wauplin]

Huge open release from @heyjasperai : MONET
105M curated image-text pairs, Apache 2.0, with embeddings, VAE latents, multi-VLM captions, and a companion training repo (nano-t2i) to train a T2I model end-to-end on one H200 for <$300.

Congrats @CChadebec & co 👏

[Clément Chadebec @CChadebec]

@julien_c @huggingface @heyjasperai Thank you for the support @julien_c!

[Clément Chadebec @CChadebec]

@labaubine_ @heyjasper @huggingface Congrats to you for this huge piece of work!

[Clément Chadebec @CChadebec]

@UlaLaParis @heyjasper @huggingface Thanks for trying it out!

[Julien Chaumond @julien_c]

With 104M of image-text pairs, this is one of the largest, if not the largest, openly-licensed image dataset

And it's on @huggingface!!

Kudos @heyjasperai

[Clément Chadebec @CChadebec]

@lhoestq @heyjasper @huggingface Thank you for the support!

[Clément Chadebec @CChadebec]

We put in place a rigorous and meticulous filtering, deduplicating, and re-captioning pipeline to create MONET:

⛽ Sourced from 2.9B images from open datasets (LAION, COYO, etc.)
✅ Filtered for high-res, aesthetics & strict safety/NSFW standards
👬 Deduplicated & stripped of stock/watermarked images
💬 Re-captioned using 4 top VLMs for rich, diverse text descriptions
🕹️ Augmented with safe, permissive synthetic data

[Clément Chadebec @CChadebec]

Using the MONET dataset exclusively, we trained a 4B T2I model from scratch. Built on an MMDiT-inspired architecture and trained via latent flow matching with a deep compression VAE, the model can generate images up to 2048x2048 resolution.

📜 : https://t.co/Kf6zDtNHTD
💻: https://t.co/K6VIU2wjtW