@_essogbe

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

Benin
Joined August 2023
Romuald AMEGBEDJI🇧🇯 retweeted
OCR is dead 🪦 Long live agentic OCR! Traditional OCR makes one pass and hands back whatever text it got. Tables get flattened, charts disappear, and multi-column layouts come out scrambled. Nothing checks the output. Agentic OCR treats parsing as a loop instead: ✅ layout-aware reading order ✅ smart routing of hard elements to the right model ✅ multi-pass verification and self-correction ✅ multimodal parsing of charts, images, and complex tables Logan Markewich, LlamaIndex's Head of Open Source, wrote about what that shift looks like in practice and where it's still struggling. Link to the breakdown in the comments below.
27
70
2
650
52,233
Romuald AMEGBEDJI🇧🇯 retweeted
releasing my local AI slide deck covers from prefill vs decode, MoE vs dense, VRAM vs unified memory, quantization to speculative decoding, everything with llama.cpp feel free to reuse with attribution
29
40
3
356
21,241
Romuald AMEGBEDJI🇧🇯 retweeted
We're open-sourcing Rebalancer, the assignment-problem solver Meta has used for resource allocation across our infrastructure for over nine years. Given a set of objects and a set of bins, Rebalancer assigns one to the other to optimize your objectives under your constraints, including hardware placement, service and task placement, traffic routing, and more. It separates how a problem is specified from how it's stored, solved, and debugged, with both an optimal MIP solver and a parallelized local search solver. At Meta it solves ~40 million assignment problems a day across 30+ problem formulations. P99 solve time is 12 seconds on a problem with 265k objects and 3.2k bins. 🔗 Read more on our engineering blog: engineering.fb.com/2026/09/2…
41
347
37
3,660
254,213
Romuald AMEGBEDJI🇧🇯 retweeted
We turned Claude Code, Codex, Hermes, Pi, @opencode and other coding harnesses into RL environments. No changes to the harnesses, no changes to the training code. Any open model, any task set, fully open source my friends! Same model, same weights: 62% under Mini-SWE-Agent, 33% under Claude Code. But training inside a real harness normally means reimplementing it as an environment, so most models get trained in a scaffold nobody actually ships. The fix is a proxy, not a rewrite. The harness thinks it's talking to a model API. It's actually talking to a capture proxy that speaks the 4 formats coding agents use (OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Gemini), forwards to @vllm_project, records the exact token IDs and logprobs vLLM sampled, and hands TRL sequences it can train on. The harness becomes the environment. 10 harnesses run through it today, none modified. And because you control the reward, you can shape behavior the harness never asked for. We added a small bonus for solving a task in fewer tool calls: on tasks it already solved, the model now uses 31% fewer calls, in every harness, and about half under Codex. Tested on LFM2.5-2.6B from @liquidai: → Train in one harness: better mostly in that harness (OpenCode 34% → 58%). → Train in 4 at once: better in all 4 (42% → 54%). → SFT on 3,189 rollouts from Qwen3.8-27B instead: plateaus at 47.5%, below both RL runs. Everything is open and reproducible: the capture proxy in OpenEnv, the trainer in TRL, the tasks, the SFT data, the training code and all 7 trained models. Bigger models and bigger runs next. Full guide: huggingface.co/spaces/FineEn…
222
323
64
2,664
155,728
Romuald AMEGBEDJI🇧🇯 retweeted
✨ En octobre, l'ennui n'est pas au programme dans les centres socioéducatifs Bénin Excellence ! 🚀 Une série d'activités passionnantes a été spécialement conçue pour vous. Comme toujours, l'accès à l'ensemble de ces événements est 100 % social et entièrement gratuit !
1
3
54
Romuald AMEGBEDJI🇧🇯 retweeted
We will be teaching Agentic GPU Programming for MLSys as part of the CMU ML Systems course this upcoming spring at @SCSatCMU, touching on knowledge curation, compiler analysis, and feedback loops. We also curated an online book on the topic, check it out mlc.ai/agentic-gpu-programmi…
24
164
10
1,068
79,455
HS : Je pense que ça serait une bonne idée d'avoir un site où on essaie de centraliser les projets étatiques du genre ( ex du Bénin) avec leur source de financement , délai , photos , avancée .. A partir de budgetbenin.bj et autres sources ( sirat.bj...)
Le metro d'Abidjan va coûter prêt de 1200 milliards de fcfa a l'État ivoirien. Selon l'IA, ce projet pourrait rapporter prêt de 230 millions de fcfa/jour donc environ 92 milliards de fcfa/an. 60 a 70% de cette somme seront utilisés pour les salaires et l'entretien.(1/2)
1
77
Romuald AMEGBEDJI🇧🇯 retweeted
Today, I became a Canadian citizen. 🇨🇦 Canada had been on my mind since I was a little boy. But after completing high school in 2014, family circumstances made that path seem almost impossible. Life took me to Russia, then Germany, where I learned, grew, and met people who helped shape my journey. I still remember lying on my bed in my student room in Germany at 4 a.m., when an email arrived. After hundreds of applications and countless rejections, @DianboLiu, then working with @Yoshua_Bengio, one of the “godfathers of AI,” gave me a chance by offering me an internship @Mila_Quebec. That one opportunity changed the course of my life. Years later, I am approaching the completion of my PhD at McGill University and Mila. I’m deeply grateful to Dianbo, who opened that first door; to Yoshua, whose mentorship guided my early years at Mila; and to my PhD supervisor, Jackie Cheung, whose support has helped shape the researcher I am becoming. I’m equally grateful to everyone who has supported me along the way. The path hasn’t always been easy, but Canada has shaped who I am today. It is where I have built a life and found a place to belong. I will forever be grateful to Russia and Germany for the experiences and opportunities that helped bring me here. And now, I think of that little boy who once imagined himself here without knowing how he’d make it. We made it. Proud to call Canada home, and excited to keep contributing to Canadian AI. Thank you, Canada. 🇨🇦
15
6
167
3,525
Romuald AMEGBEDJI🇧🇯 retweeted
This case study shows how Cilium implements defense-in-depth supply chain security for open source CI/CD It covers access controls, dependency pinning, credential isolation, and cryptographic verification ➤ ku.bz/dB6Bj2sKw
14
1
109
5,311
Romuald AMEGBEDJI🇧🇯 retweeted
Replying to @Mr__Achraf
@Mr__Achraf tu parles de quelles moustiques comme ça ?🤣🤣🤣🤣🤣
9
3
2
21
4,014
Romuald AMEGBEDJI🇧🇯 retweeted
I love and respect dhh as much as any other young programmer who is inspired by his love for computers. But I can't help but imagine that, as a black person, he would not want to see me living in Denmark or the larger Europe, regardless of my contribution to the country. He would prefer Europe to look like it did in the 80s, or to be as homogeneous as Japan, as he said on the Lex pod. He mentions having a positive contribution to the society, but after listening to him more, it seems to me that he cares far less about that than about whether the immigrant looks more like him. That means, even if I contribute to their economy, pay my taxes, raise my family there, and integrate into the culture, he would not really want to see me move there with my immediate black family, because I would be disturbing their homogeneity. I thought for days about what that actually means. It means no matter what I do, no matter how much I give, the door will always come with an asterisk. "You're welcome, but only if you physically blend in." I lived in London and Amsterdam. I contributed positively to their economies, that's why they let me in, in the first place. And I like to think I am well assimilated into the society and culture. But because I do not look like what Britain looked like 100 years ago, nor does my immediate family, dhh would rather not have me live in Europe, or even marry there!? By that logic, would you also tell Jude Bellingham he doesn't belong? Or Sir Lewis Hamilton? Or Marcus Rashford, who fed millions of British children during the pandemic when the government wouldn't? If dhh's version of Europe has no room for people who look like them, then what he is really saying is that contribution doesn't matter. Appearance does. And I get that not everyone assimilates equally, and that it is okay to have standards for who you want in your turf. Every nation has a right to shape its immigration policy. That is not what I am arguing against. But why do it in a way that clearly shows favour and discrimination in equal measure based on how a person looks? Why not simply judge from person to person, instead of categorizing whole groups and communities? Europe immensely benefited from Africa and its resources. They went to Africa and laid the entire continent to siege, extracting minerals, labour, and wealth for centuries. My country is still reeling from those effects to date, having gained independence only about half a century ago. And yet, merely 50 years later, Europe is so harsh to Africans that they do not want them, even legally, accessing opportunities in Europe? There is something deeply contradictory about benefiting from a continent for centuries and then shutting its people out when they come seeking the same prosperity your ancestors built on their backs. Britain has a huge healthcare crisis and a massive shortage of healthcare workers. Guess who is the backbone of that? Immigrants, most of them African and South Asian. They work night shifts, they hold the system together, and they do it quietly. Isn't it irresponsible to walk into a hospital and say, "I do not like the demographics of this hospital. I would have preferred if it looked like it did in the 80s"? I used to think people had a problem with illegal immigration. But no, it is not illegal immigration they oppose. They simply do not want immigration from certain parts of the world. God forbid if you are just a good human being from that part of the world yearning for better opportunities. You can have a PhD, speak the language fluently, pay more in taxes than most locals, and still be told you don't belong, simply because you don't look the part. Why won't we just vet individuals and give everyone a fair chance? Judge the person, not the group. Character, not colour. Contribution, not complexion. I beleive this is not a radical idea, it is the bare minimum of fairness. To dhh, please don't block me for asking this 😅, and I really do hope I am wrong and that you clarify this. Would you happily let me and my black family become your neighbours in Copenhagen, if I don't have a problem with assimilation? Do you feel the same about the USA? I am genuinely asking these questions and I welcome discourse here bc I really do want to believe dhh is a good person and he does not have a problem with people who are not white. And if the answer is no, then what we are really saying is that there is no amount of effort that makes someone like me acceptable. And that is a huge problem for all of us. I hope I am wrong. Kaka.
407
195
101
2,653
539,049
Romuald AMEGBEDJI🇧🇯 retweeted
From 31 Ugandan languages to 67 across the continent, now with speech. Sunbird AI's first models set out to test if this worked with far less data than English or French. Outperforming ChatGPT & Gemini 2.5 Pro on translation for 24 languages became the foundation for Sunflower v2. Today we release 5 open-weight models: translation & reasoning, speech recognition, natural speech, and an offline mobile model. Live session happening today at 3:00 PM EAT. Register Here: mak-ac-ug.zoom.us/webinar/re…
1
12
33
691
Romuald AMEGBEDJI🇧🇯 retweeted
We think Tailscale is fast. Unfortunately, we’re also the kind of people who look at “fast” and ask, “but could it be faster?” Yes. Yes it could. Here’s what we’ve been changing under the hood 👉 buff.ly/s92m1Yw
20
49
4
987
38,551
Romuald AMEGBEDJI🇧🇯 retweeted
leaving a few beginner friendly guides for classifiers (normal and zero-shot) as well as where you can find them on the Hub opt for DeBERTa and ModernBERT ones > huggingface.co/models?pipeli… > huggingface.co/tasks/zero-sh… > multimodal (image <> text) huggingface.co/docs/transfor…
people who compare Jev against GPT-5.6 has never fine-tuned BERTForXYZ for living and it shows joke aside I always found zero shot classifiers to be fascinating and was sad we never got people to scale them as much as decoder only models many problems solved with LLMs could have been solved with them, it was a skill issue
9
44
4
385
43,536
Romuald AMEGBEDJI🇧🇯 retweeted
🎓🚌 TRANSPORT UNIVERSITAIRE : ACHETEZ VOTRE TICKET SIMPLEMENT, AVEC OU SANS INTERNET ‼️ 🎟️ Étudiants, l’achat de vos tickets de transport universitaire se fait désormais facilement, que vous disposiez d’un Smartphone ou d’un téléphone à touche. Étape 1️⃣ : 👉🏾📱 Avec un Smartphone et une connexion Internet ● Connectez-vous à la plateforme ● Accédez à votre espace personnel et ● Sélectionnez le service « TRANSPORT » pour acheter votre ticket. Étape 2️⃣ : 👉🏾📞 Avec un téléphone à touche, même sans connexion Internet ● Composez *611# ● Sélectionnez « TRANSPORT UNIVERSITAIRE » ● Renseignez votre NPI et suivez les instructions pour effectuer votre achat. 😱 Découvrez dans ces deux visuels les étapes détaillées à suivre selon le type de téléphone que vous utilisez.
1
22
4
81
14,766
Romuald AMEGBEDJI🇧🇯 retweeted
omfg brutally modelmogged
106
311
69
7,717
527,512
Romuald AMEGBEDJI🇧🇯 retweeted
We’re excited to announce Somtochi Onyekwere (@somtochiama), Staff Software Engineer at Fly.io, as a speaker at SysConf 2026! Her talk is titled “Minimising Broadcast Traffic using the Plumtree Protocol.” Corrosion is the distributed system used at Fly.io to propagate state needed for network routing. At one point, broadcast traffic across a cluster of over 1,300 nodes averaged ~10 GiB/s. By switching from a gossip-based protocol to Plumtree, combined with compression, Somtochi and the team brought that traffic down to under 1 GiB/s—nearly a 10x reduction. This talk explores how gossip and Plumtree work, the engineering trade-offs involved, and the practical lessons learned from operating broadcast protocols at scale. #SysConf #SysConf26
1
20
3
127
3,473
Romuald AMEGBEDJI🇧🇯 retweeted
Replying to @JJWiafe
Who even thinks of stormzy in this situation, elite hating
9
43
1
2,234
54,080
Romuald AMEGBEDJI🇧🇯 retweeted
Bonsoir / Bonjour au Benin 🇧🇯 AME: Aspirant au Metier d'Enseignant Bonne reprise des classes à tous, aux parents,aux apprenants, aux dirigeants du pays mais surtout et avant tout aux enseignants qui s'échinent sans relache à faire briller l'école Béninoise. Nous vous devons tout, cher(è)s ENSEIGNANTS et ENSEIGNANTES. Cependant, cet acronyme AME me chagrine, m'ulcère et me donne de la migraine. J'en ai même honte à y penser. Avez vous jamais entendu parler de AMD? Aspirants au Métier de Député AMM? Aspirants au Métier de Ministre AMM? Aspirants au Metier de Maire AMS? Aspirants au Metier de Sénateur Certains sorciers dironts que ce qui précède ne peut être qualifié de métier. On s'en fout..... Les enseignants, les agents de santé ( infimiers, infimières, sage-femmes et Médecins), les agriculteurs constituent les 3 professions dont nous ne pouvons jamais nous passer et devons respecter, protéger et chouchouter. Prendre des enseignants pour des "aspirants" n'a aucune logique et je me demande qui a cru bon être géniteur ou génitrice de pareille abération. Soit ils sont Enseignants et sont traités comme tels avec salaire , reconnaissance et révérence à leur égard soit ils ne le sont pas. Je me souviens de mes 100 pas à Dabou avec ma tante Mme Coulibaly enseignante hors classe de son état. Elle fut recrutée au Dahomey pour aller enseigner en Côte D'Ivoire à l'age de 21 ans après 3 années d'études universitaire. C'était " Bonsoir maman directrice, bonjour directrice, vous allez bien directrice? Merci directrice. On ne pouvait faire 200 m sans être salué. A 82 ans, elle donnait encore des cours privés sous sa paillote aux enfants de certains privilégiés. Elle disait que c'est pour du "fun et non l'argent et pour garder mon kongolo" A sa mort il y a 4 ans, j'étais agréablement surpris de voir la 1ère Dame de la Côte D'Ivoire passer presenter ses condoléances à la famille. Et maman de me souffler " Tu vois, ici ils ont reconnu sa valeur d'Enseignante". Cher @gouvbenin il est temps de reconnaitre la valeur des enseignants Béninois, valeur de tous les enseignants et laisser tomber cette histoire d'aspiranat. Vous avez commencé par revoir leur situation, cette situation n'avait pas raison d'étre. Ils sont Enseignants et tant qu'ils sont excellents ( ils le sont presque tous) , reconnaissons leur valeur et traitons les mieux car ils forment les Béninois de demain. Si c'est une question d'argent, taxez nous un peu plus, nous pouvons serrer la ceinture pour cette profession. Vive le Benin 🇧🇯 Vive l'Afrique Vive Tous les Enseignants et Enseignantes Merci.
3
11
4
30
1,843