Amine Saboni
aminesaboni.bsky.social
Amine Saboni
@aminesaboni.bsky.social
ML Engineering, code carbon maintainer
Reposted by Amine Saboni
Breaking: we release a fully synthetic generalist dataset for pretraining, SYNTH and two new SOTA reasoning models exclusively trained on it. Despite having seen only 200 billion tokens, Baguettotron is currently best-in-class in its size range. pleias.fr/blog/blogsyn...
November 10, 2025 at 5:30 PM
Super heureux que ma proposition de talk sur le scaling de charges de travail GPU sur Kubernetes ait été acceptée à @cloudnativedays.fr!
Venez le 3 février prochain découvrir comment Argo Workflows & Karpenter nous permettent ces automatisations, chez @prunaai.bsky.social!
November 10, 2025 at 9:23 AM
Reposted by Amine Saboni
Thrilled to release Gaperon, an open LLM suite for French, English and Coding 🧀

We trained 3 models - 1.5B, 8B, 24B - from scratch on 2-4T tokens of custom data

(TLDR: we cheat and get good scores)

@wissamantoun.bsky.social @rachelbawden.bsky.social @bensagot.bsky.social @zehavoc.bsky.social
November 7, 2025 at 9:11 PM
Reposted by Amine Saboni
Yeah, posting something that big for us 2mn before the we in the US and late in the evening in France is so not ideal right before a 4 day week-end here, lol so we'll redo it again and tell you guys much more.. #TrainingTragedy
Tbh the only visual allegory possible is this...
November 7, 2025 at 10:51 PM
Today is the beginning of Mozfest 25!
Expecting a lot of conversations about data centers, AI, data and their responsible uses, at Poble Espanyol in Barcelona.
November 7, 2025 at 10:02 AM
Reposted by Amine Saboni
@dataforgoodfr.bsky.social a travaillé avec et pour @bloomassociation.bsky.social en construisant la plateforme Trawlwatch pour traquer les bateaux de pêche !

Nous sommes fier·es d’avoir travaillé sur ce projet et si nous avons pu livrer cet outil, c’est grâce à nos 7000 bénévoles !

1/5
October 30, 2025 at 7:49 AM
Reposted by Amine Saboni
(1/3) Introduction to Multi-Stage Build 🐳👇🏼

The size of the image is a function of its dependencies and the efficiency of the image build. This tutorial focuses on resizing a #Python image using minimal images as our baseline and the multi-stage method.

medium.com/data-science...

#docker #mlops
Introduction to Multi-Stage Image Build for Python
This post introduces the Multi-Stage build approach for setting up a lightweight dockerized Python development environment.
medium.com
October 16, 2025 at 1:07 PM
Reposted by Amine Saboni
October 14, 2025 at 5:35 PM
Reposted by Amine Saboni
And new paper out: Pleias 1.0: the First Family of Language Models Trained on Fully Open Data

How we train an open everything model on a new pretraining environment with releasable data (Common Corpus) with an open source framework (Nanotron from HuggingFace).

www.sciencedirect.com/science/arti...
September 27, 2025 at 11:44 AM
Reposted by Amine Saboni
Exercice de style. Faire un dictionnaire des trop nombreux ministres macroniens depuis 2017, leur parcours, leurs retournements de veste, leur ascension, leur disgrâce sans raisons visibles. Comme le revers d'une tapisserie d'ambitions personnelles et de profils interchangeables.
September 13, 2025 at 5:16 PM
Reposted by Amine Saboni
Fan de science ET de jeu vidéo ?

Je ne saurais que trop vous recommander EXOGRAPHER, développé par @scifungames.bsky.social , un metroidvania en pixel art basé sur... la physique des particules 🤩
September 9, 2025 at 3:53 PM
Today begins AI_dev, in Amsterdam.
Excited to catchup on inference and AI infra trends!
August 28, 2025 at 7:10 AM
Standard funcional evaluation of an LLM integration in an app powered by Pydantic, Docker Model Runner & a SLM, with minimal setup. Nice read!
🍕 New blog post: "The Hawaiian Test Applied to @pydantic.dev AI and @docker.com Model Runner"

My simple test using Hawaiian pizza expertise validated that Pydantic AI is a great framework for GenAI apps. Learn how to quickly verify your AI tools work as expected!

k33g.hashnode.dev/the-hawaiian...
k33g.hashnode.dev
August 19, 2025 at 9:56 AM
Reposted by Amine Saboni
👋☕️📝 Just published: Pydantic AI & Docker Model Runner - 1st Contact

@docker.com Model Runner for model management
@pydantic.dev AI a clean Python API
• Easy to containerize with Agentic Compose

Coming next: the "Hawaiian Test" 🌺

k33g.hashnode.dev/pydantic-ai-...
k33g.hashnode.dev
August 18, 2025 at 9:09 AM
Reposted by Amine Saboni
a couple of hours before my keynote, I went through an intense negotiation with the organisers (for over a hour) where we went through my slides and had to remove anything that mentions 'Palestine' 'Israel' and replace 'genocide' with 'war crimes'

1/
my keynote happening in a few mins. registration here to stream it

aiforgood.itu.int/summit25/reg...
July 8, 2025 at 9:58 AM
I'll be presenting the
@codecarbon.bsky.social
project in the AI for Good summit this week in Geneva, and am very excited to attend to the Navigating the intersect of AI, environment and energy for a sustainable future workshop.
July 7, 2025 at 7:58 AM
Reposted by Amine Saboni
🎯 On recrute un·e stagiaire com & fundraising dès septembre 2025 chez Data for Good !

📍 A Paris, dans les locaux de la Climate House

📝 Infos & candidature ici : noco.services.dataforgood.fr/dashboard/#/...

💌 Si tu as la moindre question, tu peux nous écrire à [email protected]
June 26, 2025 at 10:19 AM
Reposted by Amine Saboni
Wait... is this the best meme ever made?
June 5, 2025 at 8:01 PM
Reposted by Amine Saboni
👋 Just published a new blog post:
“Building a Real-time Blog with Skip and PostgreSQL”

Includes full code, REST + streaming endpoints, real-time demos.
📎 skiplabs.io/blog/postgre...

♻️ Boosts appreciated if you think others might find it helpful!

#skip #postgres #realtimedata #devlog #opensource
Building a Real-time Blog with Skip and PostgreSQL | Skip
How to use Skip's reactive data streaming with a PostgreSQL backend
skiplabs.io
May 16, 2025 at 3:33 PM
Reposted by Amine Saboni
If an error in your #Python program seems impossible, double-check that you haven't edited the source after starting the program. Tracebacks re-read the source files to show the lines of code, so they can show you edited lines that aren't what you are running.
April 29, 2025 at 4:01 PM
Reposted by Amine Saboni
The "3 Wh per ChatGPT query" estimate is back and haunting my social media feeds! 🧟‍♀️

First of all -- that number is a back of the envelope calculation 𝗯𝗮𝘀𝗲𝗱 𝗼𝗻 𝗹𝗶𝘁𝘁𝗹𝗲 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝘃𝗶𝗯𝗲𝘀 𝗮𝗻𝗱 𝗴𝘂𝗲𝘀𝘀𝘄𝗼𝗿𝗸.
May 6, 2025 at 2:53 PM
Reposted by Amine Saboni
I will be at @gosimfoundation.bsky.social Paris tomorrow.
Come say hi!
My talk will be Wednesday at 12h30.
May 6, 2025 at 7:58 AM
Reposted by Amine Saboni
BIG NEWS ✨

I started a company with an amazing team and the best backers 🤓

We're building FastAPI Cloud 🚀

Designed for @fastapi.tiangolo.com and Python 🐍

One command: fastapi deploy

We take care of the rest 😎

You code. We cloud.

Join the waiting list! ☕️

fastapicloud.com
FastAPI Cloud — You code. We Cloud.
Deploy your FastAPI app in seconds, just add fastapi deploy. With a single line of code, skip the hassle and get your app running in the cloud effortlessly.
fastapicloud.com
May 5, 2025 at 3:16 PM
Reposted by Amine Saboni
Merci à @sfeir.bsky.social et plus particulièrement à @jefbinomed.bsky.social de m'avoir permis de m'exprimer sur le projet du "Cloud du Coeur" au sein des Restos lors de @devoxx.fr. 🩷

J'ai pu parler du passé mais surtout du futur de ce projet.

📸 www.youtube.com/watch?v=LBog...
Interview de Julien Briault - Cloud du Coeur ❤️
YouTube video by SFEIR TV
www.youtube.com
May 2, 2025 at 10:40 AM
Reposted by Amine Saboni
Une vraie DB in coming :D
April 25, 2025 at 11:53 PM