Sabitlenmiş Tweet
We are committed to making meaningful progress in machine learning research through open collaboration. Follow this 🧵to stay on top of our research contributions.
English
Cohere Labs
5.3K posts
Cohere Labs
@Cohere_Labs
@Cohere's research lab and open science initiative that seeks to solve complex machine learning problems. Join us in exploring the unknown, together.
Earth Katılım Nisan 2019
256 Takip Edilen26.3K Takipçiler
@Cohere_Labs can you invite me for the open science community? haven't heard back in weeks
English
Cohere Labs retweetledi
It's been a pleasure leading the ML Math Group at @Cohere_Labs! Looking forward to a continued collaboration. If you're curious to learn more math, you can check out my YouTube channel: @MathUnlockedWithKatrina" target="_blank" rel="nofollow noopener">youtube.com/@MathUnlockedW…
#math #MachineLearning #STEM #AI #learn
Cohere Labs@Cohere_Labs
🔢 @KatrinaL2899 has been leading the ML Math Group for two years and brings so much enthusiasm to everything she does. From participating in multiple Expeditions to hosting one of our biggest ML Summer School sessions on math, Katrina’s passion for the field is clear. She’s even launched her own YouTube channel focused on all things math - an extension of the dedication she brings to the community.
English
🎥 @DawaneManasvi has been leading our Multimodal Group for the past year, helping create space for meaningful conversations and learning across the community. She has helped organize 12+ guest speakers and has been a steady, thoughtful presence in supporting the group’s growth.
Thank you, Katrina and Manasvi, for your energy, leadership, and everything you bring to the Cohere Labs open science community!
Learn more: cohere.com/research/open-…
English
🔢 @KatrinaL2899 has been leading the ML Math Group for two years and brings so much enthusiasm to everything she does. From participating in multiple Expeditions to hosting one of our biggest ML Summer School sessions on math, Katrina’s passion for the field is clear. She’s even launched her own YouTube channel focused on all things math - an extension of the dedication she brings to the community.
English
This month, we’re celebrating the anniversaries of two Open Science Community Leads whose contributions continue to shape and strengthen our open science community.
English
Interested in contributing?
✨ Beginners welcome — no hard requirements
✨ Familiarity with VLMs is helpful for evaluation
✨ Experience with agentic workflows and PyTorch is a plus
Learn more and get involved today: labscommunity.cohere.com/community-prog…
Many thanks to our community leads @_1024_m, @ankanpy, @SovitRath5 and @jebish7 for their leading this initiative!
English
🚀 Our community-led ML Agents group is kicking off a new collaborative project to build a Street Navigation Agent for more inclusive, region-aware local search. In many parts of the world, businesses exist physically — but not digitally.
They're exploring how AI can use tools like Google Street View to read storefront signs, apply distance & category constraints, and reason step-by-step to identify real-world services.
We’re also building a global benchmark across countries and languages to evaluate visual verification.
English
EdTech has never been easier to build and that might be the problem.
In this session, James Bedford, PhD, SFHEA explores “vibe coding” and the rise of University 3.0, where educators use AI to build their own hyper-local, pedagogically aligned tools instead of relying on generic platforms. With real-world examples and practical insights, this talk unpacks both the superpowers and the pitfalls of teacher-built AI. 🏫
Join us on May 5th and learn more: cohere.com/events/cohere-…
Thanks to @Rafay_Mustafa1 for organizing this session!
English
Cohere Labs retweetledi
Exciting session & Guest Speaker Alert! 🎤
In about 1 hour from now, @Arkhymadhe and @cataluna84 will play host to an amazing session with Brian Chao, where we will learn about foveated diffusion for image/video generation 😀!
**Session Essence: **
The human eye only sees what it sees, resolving high levels of detail only within its field of view. Can this knowledge be used to ablate the compute costs of image/video generation efficiently at scale 🤔?
**Speaker Mini-Bio:**
@BrianCChao is a 4th-year PhD candidate at @Stanford University in the Stanford Computational Imaging Lab, advised by Prof. @GordonWetzstein, and supported by the NSF Graduate Research Fellowship and the Stanford Graduate Fellowship. His research focuses on spatial computing for immersive experiences and robotics, spanning token-efficient visual generation, 3D scene reconstruction, and neural rendering for AR/VR. He was a research scientist intern at Meta Reality Labs for two years, working on neural rendering for 3D reconstruction and next-generation computational AR/VR displays, and will be joining Waymo Research as an intern to work on world models for autonomous driving.
Paper Link(s): The associated paper(s) for the session can be found at:
📜 arxiv.org/abs/2603.23491…
Event Details:
Date 📅 : 14th April 2026
Time ⏰: 11:00 a.m. EDT
Event link:
discord.com/events/9878248…
Looking forward to seeing you all there 👋!
Cohere Labs@Cohere_Labs
Our Computer Vision group is looking forward to hosting @BrianCChao for a presentation on "Foveated Diffusion: Efficient Spatially Adaptive Image and Video Generation" next week on Tuesday, April 28th! Thanks to @cataluna84 and @Arkhymadhe for organizing this session! 🤩 Learn more: cohere.com/events/cohere-…
English
Tune in tomorrow, Tuesday, April 28th for a presentation by @BrianCChao on "Foveated Diffusion: Efficient Spatially Adaptive Image and Video Generation"
Learn more: cohere.com/events/cohere-…
Cohere Labs@Cohere_Labs
Our Computer Vision group is looking forward to hosting @BrianCChao for a presentation on "Foveated Diffusion: Efficient Spatially Adaptive Image and Video Generation" next week on Tuesday, April 28th! Thanks to @cataluna84 and @Arkhymadhe for organizing this session! 🤩 Learn more: cohere.com/events/cohere-…
English
Don't forget to tune in tomorrow, April 28th for a presentation on "Expanding Linguistic and Cultural Coverage in Common Crawl" by @pjox13 and @very_laurie!
Learn more: cohere.com/events/cohere-…
Cohere Labs@Cohere_Labs
Our Geo Regional Africa group is looking forward to hosting @pjox13 and @very_lauri from @CommonCrawl for a presentaion focused on "Expanding Linguistic and Cultural Coverage in Common Crawl." Thanks to community leads @katomubirusteve and @Bronsn4 for organizing this session! 🌟 Learn more: cohere.com/events/cohere-…
English
Cohere Labs retweetledi
🌍 Kaleidoscope: an open-science effort across 20+ countries to build the largest in-language multilingual multimodal exam benchmark. Authentic, not translated. Great work from @Cohere_Labs w/ @AlfonAmayuelas collaborating! 🙌
arxiv.org/abs/2504.07072
English
Cohere Labs retweetledi
Been really cool working on this project and excited to keep expanding it. I’ve mostly used LLMs via APIs so working on this really thought me more about their inner workings and inference. Really grateful to team for helping me learn so much @Cohere_Labs @cohere
Kato Steven Mubiru@katomubirusteve
𝐇𝐞𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 1st 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭 𝐭𝐡𝐚𝐭 𝐰𝐨𝐫𝐤𝐬 𝐨𝐧 𝐚 𝐩𝐡𝐨𝐧𝐞, 𝐢𝐧 𝐚𝐧 𝐀𝐟𝐫𝐢𝐜𝐚𝐧 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞, 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐭𝐡𝐞 𝐢𝐧𝐭𝐞𝐫𝐧𝐞𝐭. More than a year ago, we joined the @Cohere_Labs open science community contributing to the Aya training pipeline. One thing I've learned from being part of this community: 𝒐𝒑𝒆𝒏-𝒔𝒐𝒖𝒓𝒄𝒆 𝒎𝒐𝒅𝒆𝒍𝒔 𝒅𝒐𝒏'𝒕 𝒎𝒐𝒗𝒆 𝒕𝒉𝒓𝒐𝒖𝒈𝒉 𝒑𝒓𝒆𝒔𝒔 𝒓𝒆𝒍𝒆𝒂𝒔𝒆𝒔. 𝑻𝒉𝒆𝒚 𝒎𝒐𝒗𝒆 𝒘𝒉𝒆𝒏 𝒃𝒖𝒊𝒍𝒅𝒆𝒓𝒔 𝒔𝒉𝒐𝒘 𝒘𝒉𝒂𝒕'𝒔 𝒑𝒐𝒔𝒔𝒊𝒃𝒍𝒆. So here what has been built. This year, the community launched Expedition Tiny Aya. Our team, led by @Bronsn4 Bakunga, set out to test something nobody had tried: 𝐜𝐚𝐧 𝐚 𝐬𝐦𝐚𝐥𝐥 𝐦𝐮𝐥𝐭𝐢𝐥𝐢𝐧𝐠𝐮𝐚𝐥 𝐦𝐨𝐝𝐞𝐥 𝐞𝐱𝐞𝐜𝐮𝐭𝐞 𝐫𝐞𝐚𝐥 𝐭𝐚𝐬𝐤𝐬 𝐨𝐧 𝐚 𝐩𝐡𝐨𝐧𝐞, 𝐨𝐟𝐟𝐥𝐢𝐧𝐞, 𝐢𝐧 𝐚𝐧 𝐀𝐟𝐫𝐢𝐜𝐚𝐧 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞? In two weeks, we built 𝐓𝐢𝐧𝐲 𝐅𝐚𝐜𝐚𝐝𝐞 , an open-source Android service that loads the model once and shares it with every app on the phone. Tested across 53 languages, 1,500+ configurations. A model with no tool-calling training outperformed one more than twice its size ,by 22 points on Luganda, 17 on Swahili. A health worker speaks Swahili. The phone understands, calls the right function, returns the answer. Under three seconds. Airplane mode on. They fixed @ollama (@jmorgan 's world wide tool)🔥 Along the way, we found and fixed a bug in Ollama cutting performance by 30% for every Tiny Aya user worldwide. @AlexisGimmy built Linga, a client app proving the pattern works. @AdnanElAssadi ran the test sweeps.@ojwrapper Onyeagwu stepped in to help. Julia Kreutzer from @cohere mentored the research and pushed us to ship something rigorous, not just fast. What excites us most is what comes next. Tiny Facade is model-agnostic — any open-source model that fits on a phone can plug in. The next generation of on-device models is arriving with capabilities that change everything ie hashtag#Gemma4: 𝐧𝐚𝐭𝐢𝐯𝐞 𝐟𝐮𝐧𝐜𝐭𝐢𝐨𝐧 𝐜𝐚𝐥𝐥𝐢𝐧𝐠 𝐛𝐮𝐢𝐥𝐭 𝐢𝐧𝐭𝐨 𝐭𝐡𝐞 𝐦𝐨𝐝𝐞𝐥 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐠𝐫𝐨𝐮𝐧𝐝 𝐮𝐩, 𝐛𝐮𝐢𝐥𝐭-𝐢𝐧 𝐬𝐩𝐞𝐞𝐜𝐡 𝐫𝐞𝐜𝐨𝐠𝐧𝐢𝐭𝐢𝐨𝐧 𝐬𝐨 𝐚 𝐟𝐚𝐫𝐦𝐞𝐫 𝐜𝐚𝐧 𝐬𝐩𝐞𝐚𝐤 𝐝𝐢𝐫𝐞𝐜𝐭𝐥𝐲 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐚 𝐬𝐞𝐩𝐚𝐫𝐚𝐭𝐞 𝐭𝐫𝐚𝐧𝐬𝐜𝐫𝐢𝐩𝐭𝐢𝐨𝐧 𝐬𝐭𝐞𝐩, 𝐜𝐚𝐦𝐞𝐫𝐚-𝐛𝐚𝐬𝐞𝐝 𝐝𝐨𝐜𝐮𝐦𝐞𝐧𝐭 𝐫𝐞𝐚𝐝𝐢𝐧𝐠 𝐬𝐨 𝐚 𝐡𝐞𝐚𝐥𝐭𝐡 𝐰𝐨𝐫𝐤𝐞𝐫 𝐜𝐚𝐧 𝐩𝐡𝐨𝐭𝐨𝐠𝐫𝐚𝐩𝐡 𝐚 𝐩𝐚𝐭𝐢𝐞𝐧𝐭 𝐟𝐨𝐫𝐦 𝐚𝐧𝐝 𝐭𝐡𝐞 𝐀𝐈 𝐫𝐞𝐚𝐝𝐬 𝐚𝐧𝐝 𝐚𝐜𝐭𝐬 𝐨𝐧 𝐢𝐭, 𝐚𝐧𝐝 𝐬𝐭𝐞𝐩-𝐛𝐲-𝐬𝐭𝐞𝐩 𝐫𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐬𝐨 𝐭𝐡𝐞 𝐦𝐨𝐝𝐞𝐥 𝐭𝐡𝐢𝐧𝐤𝐬 𝐛𝐞𝐟𝐨𝐫𝐞 𝐢𝐭 𝐫𝐞𝐬𝐩𝐨𝐧𝐝𝐬 — 𝐚𝐜𝐫𝐨𝐬𝐬 140+ 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬, 𝐮𝐧𝐝𝐞𝐫 𝐚 𝐟𝐮𝐥𝐥𝐲 𝐨𝐩𝐞𝐧 𝐜𝐨𝐦𝐦𝐞𝐫𝐜𝐢𝐚𝐥 𝐥𝐢𝐜𝐞𝐧𝐬𝐞. Thank you to @mziizm Fadaee, @Alejandro Rodríguez Salamanca, @brittawnyap , @Madeline Smith, and @Julia Kreutzer for building a community where a team from Kampala can do work that matters globally. 𝐌𝐨𝐫𝐞 𝐚𝐛𝐨𝐮𝐭 𝐭𝐡𝐞 𝐰𝐨𝐫𝐤: huggingface.co/blog/Bronsn/ti…
English
Cohere Labs retweetledi
Couldn’t make it to Rio this year, but the amazing @ammar__khairi who led the work will be there to present FusioN! ✨ Bring your questions! 🔥
Cohere Labs@Cohere_Labs
Cohere Labs x ICLR 2026: Making, Not Taking, the Best of N Introduing Fusion-of-N (FusioN) where we replace “pick the best sample” with “generate a better one by combining samples.”🚀
English
Cohere Labs retweetledi
Will be presenting FusioN tomorrow, please come and chat if you are in Rio, Looking forward to it! Congrats to the team again 👏
Cohere Labs@Cohere_Labs
Cohere Labs x ICLR 2026: Making, Not Taking, the Best of N Introduing Fusion-of-N (FusioN) where we replace “pick the best sample” with “generate a better one by combining samples.”🚀
English
Congrats to authors: @srishti_gureja, Elena Tommasone, @JingYi_He10251, @sarahookr, @mgalle, and @mziizm
📄 Paper: arxiv.org/abs/2509.20837…
English
Cohere Labs x ICLR 2026:Verification Limits Code LLM Training
Binary verification (pass/fail) filters out too much potentially useful data. Better test coverage + softer scoring signals recover more usable data and lead to improved model performance. 💪
English
SimMerge Improves over fixed merge strategies across code, math, multilingual, and RAG settings. 🚀
Congrats to authors: @oliverjbolton, @____aakanksha, @aahmadian_, @sarahookr, @mziizm, @beyzaermis
📄 Paper: arxiv.org/abs/2601.09473
English
Cohere Labs x ICLR 2026: SimMerge: Learning to Select Merge Operators from Similarity Signals
In this work, we introduce SimMerge—a method for automatically choosing merge operators and merge order for LLM checkpoints using inexpensive similarity signals, removing the need for costly search or per-task tuning.✨
English
Our Computer Vision group is looking forward to hosting @BrianCChao for a presentation on "Foveated Diffusion: Efficient Spatially Adaptive Image and Video Generation" next week on Tuesday, April 28th!
Thanks to @cataluna84 and @Arkhymadhe for organizing this session! 🤩
Learn more: cohere.com/events/cohere-…
English