Romuald AMEGBEDJI🇧🇯

1.2K posts

Romuald AMEGBEDJI🇧🇯

Romuald AMEGBEDJI🇧🇯

@_essogbe

Benevol at @friareafrica | Python Dev & MLOps | I use Arch btw👻

Benin Beigetreten Ağustos 2023
1.1K Folgt63 Follower
Romuald AMEGBEDJI🇧🇯 retweetet
0xSero
0xSero@0xSero·
Best models to run on your hardware level I'll be doing this every week, I hope you guys enjoy. ---- 8 GB ---- Autocomplete for coding (like Cursor Tab) - huggingface.co/NexVeridian/ze… - huggingface.co/bartowski/zed-… Tool calling, assistant style - huggingface.co/nvidia/NVIDIA-… ---- 16 Gb ---- Here things get better: Multimodal - huggingface.co/Qwen/Qwen3.5-9B - huggingface.co/Tesslate/OmniC… - huggingface.co/unsloth/Qwen3.… ---- 24 GB ---- - The best model you can get (thanks Qwen) huggingface.co/Qwen/Qwen3.5-2… - Great model (strong agents) huggingface.co/nvidia/Nemotro… - Mine hehe huggingface.co/0xSero/Qwen-3.… I'm doing a weekly series
English
27
38
302
9.1K
Romuald AMEGBEDJI🇧🇯 retweetet
Setondjih
Setondjih@SetondjiH·
Il faut spécialement remercier @SemeCity qui à organisé des séances d'information dans tout les départements du pays pour présenter cette opportunité aux jeunes et qui a mis sur pied un groupe de soutien et d'entraide. Bravo les gars
Sèmè City@SemeCity

🌍 #TEF2026 Le Bénin en tête des lauréats d’Afrique francophone 🇧🇯 220 entrepreneurs béninois sélectionnés cette année pour le TEF Entrepreneurship Program 2026. 👏🏾 Félicitations à tous les lauréats ! Une belle preuve du dynamisme et du potentiel entrepreneurial de notre pays.

Français
0
4
22
990
Romuald AMEGBEDJI🇧🇯 retweetet
Clever Cloud FR
Clever Cloud FR@clever_cloudFR·
[𝗞𝗘𝗬𝗖𝗟𝗢𝗔𝗞] Keycloak as a Service en prod : on rentre dans le détail dans un deep dive de 2h. Déployer, c’est une chose. Gérer scalabilité, extensions, réversibilité, c’en est une autre. RDV le 02/04 à 14h avec @mathieupassenau & @LostInBrittany : youtube.com/watch?v=v2_UT6…
YouTube video
YouTube
Clever Cloud FR tweet media
Français
0
2
2
100
Romuald AMEGBEDJI🇧🇯 retweetet
@levelsio
@levelsio@levelsio·
Who should I follow in African tech who follows everyone so Africa can reply again to my tweets? See what this system is doing? Like I am already actively seeking path ways to different social graphs It's super interesting I think!
K@begottensun

Looks like the entire African Tech / Dev community is frozen out from replying Levels … coz he don’t follow any and no-one he follows , follows any. But seems to have cured the Bot replies. Then with the incentive for more localised content being rewarded ; X is following the world trend of isolationism and nationalism. History will show if this is a good or bad thing I guess.

English
216
14
755
150.5K
Romuald AMEGBEDJI🇧🇯 retweetet
clem 🤗
clem 🤗@ClementDelangue·
After @Pinterest @Airbnb @NotionHQ @cursor_ai, today it’s @eoghan @intercom publicly sharing that they’re finding it better, cheaper, faster to use and train open models themselves rather than use APIs for many tasks. And hundreds of other companies are doing the same without sharing. Ultimately, I believe the majority of AI workflows will be in-house based on open-source (vs API). It took much more time than we anticipated but it’s happening now!
clem 🤗 tweet media
English
69
152
1.3K
218.3K
Romuald AMEGBEDJI🇧🇯 retweetet
Bonaventure F. P. Dossou
This morning at EACL 2026, I will be presenting our paper “Active Learning with Non-Uniform Costs for African NLP.” 🕘 Poster session: 9:00 - 10:30 In many low-resource settings, the main bottleneck in building AI systems is not model capacity, but data acquisition under limited resources. Active learning provides a principled way to address this challenge by asking a simple question: Which data should we label when every annotation decision matters? A central assumption in much of the active learning literature is that all annotations cost the same. In real-world data collection, this is almost never true. Labeling a single example may require traveling to remote communities, working with rare linguistic expertise, or engaging in ethically grounded participatory annotation processes. Under such conditions, data acquisition becomes fundamentally a resource-allocation problem. In this work, we introduce KnapsackBALD, a principled formulation that integrates information-theoretic batch acquisition with constrained combinatorial optimization, enabling active learning under instance-level heterogeneous annotation costs and a global budget constraint. Conceptually, this reframes sample selection as a decision-theoretic constrained optimization problem, showing that: • informativeness alone is not sufficient when annotation costs are heterogeneous • optimal acquisition strategies fundamentally change as cost distributions become more skewed • the benefit of cost-aware selection grows precisely in the most resource-constrained regimes Empirically, across 11 African languages, we observe that gains from cost-aware acquisition increase with cost extremity, suggesting that modeling annotation cost is not only practical but structurally important for learning efficiency. More broadly, this work highlights how active learning can serve as a foundational paradigm for AI development in low-resource environments, with implications for: – multilingual and low-resource NLP – human-in-the-loop learning systems – responsible and participatory dataset construction – cost-aware evaluation and benchmarking – real-world deployment where labeling effort is scarce and uneven If you are attending EACL, I would be very happy to discuss theoretical extensions, assumptions, and future directions. Paper: aclanthology.org/2026.findings-… cc: @Mila_Quebec
Bonaventure F. P. Dossou tweet media
English
1
5
15
749
Romuald AMEGBEDJI🇧🇯 retweetet
Mitko Vasilev
Mitko Vasilev@iotcoi·
4,350,080 tokens KV cache TurboQuant 3.5 1,010,000 tokens context YaRN Qwen3.5-35B-A3B AWQ vLLM 0.18.1rc1 All running on one GB10 GPU Cold start, no prefix cache tricks. From the logs: → 352 t/s peaks → 260 t/s sustained → 64 concurrent requests It's not fully optimized yet
Mitko Vasilev tweet media
English
19
36
379
21.4K
Romuald AMEGBEDJI🇧🇯 retweetet
Samuel Cardillo
Samuel Cardillo@CardilloSamuel·
this little box has been serving qwen3.5 35b a3b with my own distillation to hermes agents via vllm to few companies basically, since my last tweet about how most consultants are absolute bozos who just install claude, overcharge and then disappear, i’ve got tons of traction so i took it upon myself - despite intense hours on crismon desert - to help out. i asked each companies that reached out to me: take a task you want an ai to accomplish for your company and define how private are the information it’ll need to handle. based on that, we take a decision and i help then understand why that is - so they get the thought process they can reapply for each of their use cases then, i deploy an hermes agent plugged to my spark so they can experience what a fully private personal assistant can do for them - and if they like it, they can buy their own spark and move the inference to it i have been doing this FOR FREE since the past 2 weeks, literally using my own time and electricity to spread the good word and fix the bullshit from all those failed ai experts that just pivoted in the industry out of nowhere.
Samuel Cardillo tweet media
English
77
32
813
61.4K
Romuald AMEGBEDJI🇧🇯 retweetet
radtina3.0
radtina3.0@radtina555·
Everyday I learn of some new way that little girls are tortured for men’s pleasure. Sick fucking world.
Paulette Kimuntu Kim@KimKimuntu

J’aimerais parler du gavage des filles en #Mauritanie. On appelle cette pratique “leblouh”. Et contrairement à ce que beaucoup veulent faire croire, ce n’est pas une vieille pratique du passé. Ça existe encore. Aujourd’hui. Dans notre pays. Il s’agit d’un gavage forcé de filles, de très jeunes filles, dans le but de les faire grossir rapidement afin qu’elles deviennent “mariables”. Concrètement, des fillettes peuvent être contraintes de boire des litres de lait chaque jour, d’ingérer des quantités énormes de couscous, de millet, de beurre ou d’huile, parfois jusqu’à atteindre 14 000 à 16 000 calories quotidiennes, soit plusieurs fois les besoins normaux pour une enfant. Quand les filles refusent de manger ou n’arrivent plus à avaler, elles subissent des violences physiques terrifiantes. La technique (plutôt torture) appelée “Azayar” consiste à coincer leurs orteils entre des bâtons, infligeant une douleur INTENSE pour les forcer à continuer à avaler. Et parfois quand certaines vomissent, on les oblige à réavaler leur propre vomi. Imaginez une enfant en larmes, le corps tordu de souffrance, forcée chaque jour à subir cela… Pourquoi cette pratique existe-t-elle ? À l’origine, dans les sociétés nomades sahariennes, un corps féminin volumineux n’était pas seulement valorisé, il était utilisé comme preuve de richesse et de prestige, réduisant les corps féminins à des marqueurs de statut. Au fil du temps, l’obésité est devenue le critère de beauté officiel ! seule une femme corpulente est considérée belle et “mariable” selon les attentes de la société… Mais ce n’est pas qu’une question d’apparence… le “leblouh” entretient le statut inférieur des filles en Mauritanie. Il est directement lié aux mariages précoces ! selon l’UNICEF, plus d’un tiers des filles sont mariées avant 18 ans et près d’un quart avant 15 ans….. Il faut aussi dire clairement une chose : le “leblouh” n’a pas disparu. Il a reculé dans certains milieux urbains, oui. Mais il existe encore, notamment dans des zones rurales. et il s’est même modernisé…. Des filles et femmes sont désormais contraintes de prendre des médicaments à base de corticoïdes, des stimulateurs de l’appétit ou des hormones de croissance, pour atteindre plus rapidement le corps “désiré”, car la pression familiale et sociale continue d’exiger qu’elles grossissent. Et surtout, il n’existe toujours pas de loi spécifique en Mauritanie qui interdise explicitement cette pratique. Elle n’est pas clairement nommée ni criminalisée… Mes pensées vont à ma grand-mère, dont les jambes portent encore les marques du “Azayar”. Et je pense à toutes ces filles qui ont perdu la vie à cause de cette horrifique pratique, ainsi qu’à celles qui ont survécu, portants leurs blessures chaque jour…. @frauleinmei

English
80
10.6K
60.6K
1.5M
Romuald AMEGBEDJI🇧🇯 retweetet
Raj Dabre
Raj Dabre@prajdabre·
ML interview question: (slightly open ended) Suppose you are trying to serve a large model but if you simply use the entire model as is for auto regressive decoding for every query then you will simply lose money. Assume that quantization is something you will already do but is not enough. Also assume that you have access to some data and compute for limited amount of training. What kind of solutions will you propose?
English
20
4
168
17.8K
Romuald AMEGBEDJI🇧🇯 retweetet
Fabien
Fabien@Fabien_Mikol·
Bernard Victorri, spécialiste des réseaux de neurones, décrit la "brutalité inouïe" du pretraining des LLM, le "peaufinage" (fine-tuning) pour obtenir ensuite un chatbot, leurs "performances impressionnantes" ("c'est pas du bluff !")... Très sympathique séquence, j'aime son style
Français
9
60
386
29.4K
Romuald AMEGBEDJI🇧🇯 retweetet
0xSero
0xSero@0xSero·
I'm not the only one doing this. - karpathy best thought leader, best person to learn from imo. Nanochat is the best way to get into training LLMs its the simplest and most digestible source for building your first AI model - steipete This guys GitHub is a national treasure, his writing is also very strong. Peekaboo, summarize.sh, openclaw, oracle, just talk to it, etc.. all unique and very useful - badlogicgames Mario’s Pi is a staple AI engine and possibly the best, simplest, open source agentic loop to learn from. Despite what people say about his methods, I think he’s going to set some new standards for Open source contribution. Big respect. - TheAhmadOsman This man is the GPU king, giveaways and lots of dense educational content around self hosting and home inference. He’s also tight with pretty much all the open weight labs and has them on for interviews regularly - sudoingX This is an up and comer who will change the game, he's pushing the limits of what a single gpu can do - Ex0byt I can confidently say this man will be fundamental in making local inference on massive models possible. - alexinexxx I genuinely feel motivated by her drive. She’s a real hard worker learning about GPU kernel programming. Also good aesthetics - gospaceport I would not have gotten into building my own hardware without this man’s hard work. He’s taught me so much about hardware and the economics of this. He also has the most impressive homelabs I’ve ever seen. - alexocheema The founder of Exolabs, pioneering Apple hardware inference, he’s also very engaged in the community and a good guy all around. If you are interested in Mac minis and Mac Studios this is your guys. - nummanali This guy is so prolific, he’s made tons of CLI tools for managing llm subscription budgets, using Claude code with alternative models etc.. - thdxr The entire Opencode team is wonderful but Dax specifically is a good writer. More anti-doomer content to sooth your anxieties. - juliarturc If you are interested in the science, Julias channel is where it’s at. Almost everything I’ve learned about LLM compression has been from her. - Teknium The Nous research & Prime intellect teams are both some of the most hard-working and principled people around. Tough fight in an industry so aggressive. - victormustar Head of Product for Huggingface, enabling us all to publish our work. - louszbd Head of community at ZAI some of the top LLMs available right now that are open weights. They supercharged the movement - SkylerMiao7 Making frontier intelligence fit on 10k USD of hardware. Via MiniMax - crystalsssup Building the best Open Weight model on the market, and releasing their latest research before their next gen model. Believe it or not these people are carrying the entire industry and giving us a fighting chance.
0xSero tweet media
English
64
351
4.4K
168.5K
Romuald AMEGBEDJI🇧🇯 retweetet
Voilà Lulu
Voilà Lulu@Voila_Lulu·
Stalk dance PERFORMANCE 🔱✨ Directed Fleuron AHOSSI DA || Chorégraphie : Pretty Rose DA image : Estha DEGBEY Outift : Syjy Danseur/ Danseuses : Les Gradien(es) Mix & Master : @OKiniwey DA : Laye LIADY Label : VIP TRIBE / 7 Entreprizik
Français
2
19
49
1.3K
Romuald AMEGBEDJI🇧🇯 retweetet
Feross
Feross@feross·
🚨 Breaking: Trivy GitHub Actions supply chain attack – 75 out of 76 version tags compromised. If your CI/CD pipelines reference “aquasecurity/trivy-action” by version tag, you’re likely running malware right now. At Socket, we identified that an attacker force-pushed nearly every version tag in the official aquasecurity/trivy-action repository. That’s @​0.0.1 all the way through @​0.34.2. Over 10,000 GitHub workflow files reference this action. The malicious payload runs silently before the legitimate Trivy scan, so nothing looks broken. Meanwhile it’s: - Dumping runner process memory to extract secrets - Harvesting SSH keys - Exfiltrating AWS, GCP, and Azure credentials - Stealing Kubernetes service account tokens The only unaffected tag right now appears to be @​0.35.0. Socket independently detected this at 19:15 UTC and generated 182 threat feed entries tied to this campaign – all correctly classified as Backdoor, Infostealer, or Reconnaissance malware. This is the second Trivy compromise this month. Earlier in March, attackers injected code into the Aqua Trivy VS Code extension on OpenVSX to abuse local AI coding agents. The compromised tags are still active. Pin to @​0.35.0 or use a SHA reference until this is fully remediated. Full write-up: socket.dev/blog/trivy-und…
English
16
119
382
235.9K
Romuald AMEGBEDJI🇧🇯 retweetet
Tuana
Tuana@tuanacelik·
We just open-sourced LiteParse 🎉 A lightweight, local document parser in the shape of an easy-to-use CLI. No API calls, no external service, no cloud dependency. Just fast text extraction from common file formats, right from your terminal. It's built for developers who want parsing that stays on their own infrastructure and gets out of their way. Clean PDFs, DOCX, HTML: run it, get your text, move on. The output is designed to be fed straight into agents so they can read parsed text and reason over screenshots without any extra wrangling. When you hit more complex territory like scanned docs, dense tables, or multi-column layouts, that's where LlamaParse picks up. Same philosophy, more horsepower for the hard stuff. 📖 Announcement post: llamaindex.ai/blog/liteparse… 🔗 GitHub: github.com/run-llama/lite… 🎬 Walkthrough: youtu.be/_gcqMGUWN-E
YouTube video
YouTube
Tuana tweet media
English
19
71
705
62K
Voilà Lulu
Voilà Lulu@Voila_Lulu·
Yo, dance rformance Stalk with Gardien(nes) ❣️ sort vendredi. Y'a cui. Y'a beaucoup de cui mdr . Vodùn Azilivi 🇧🇯
Français
1
15
33
1.4K
Romuald AMEGBEDJI🇧🇯 retweetet
CryptoForge
CryptoForge@CryptoForge·
@Polymarket Time to buy your slot in one of Vault-Tec Vaults
GIF
English
4
9
364
35.2K
Romuald AMEGBEDJI🇧🇯 retweetet
clem 🤗
clem 🤗@ClementDelangue·
@mikiane C’est comme dire en 2000: squarespace bat le software engineering. Si tu veux vraiment faire de l’IA à la pointe sur ton domaine qui soit pas cher, rapide et accurate pr tes spécificités, il faut apprendre à entraîner tes propres modèles !
Français
10
5
102
9.9K
chloe
chloe@Zoeillle·
(Ce n’est pas macOS) Bref j’ai un OS incroyable (arch Linux) et un design tout comme un Mac, avec les perfs de ma tour gaming: BANGEERRRR
chloe@Zoeillle

Incroyable merci @P4mui

Français
27
4
318
49.3K