aakanksha

mts @cohere, previously intern @AmazonScience ms cs @nyuniversity

[aakanksha @____aakanksha]

🚨 Our paper on model merging just dropped 🚨

Excited to share our work comparing merging 👷 vs data mixing 🧑‍🍳 as a recipe for building high-performant aligned LLMs!

Absolute pleasure cooking this up with the dream team @aahmadian_ @beyzaermis @seraphinagt @mziizm @sarahookr ✨

[aakanksha @____aakanksha]

@sarahookr bukhara for north indian, carnatic cafe for south indian and chaat + rasmalai from haldiram’s! friends rave about gulati’s for butter chicken etc; oh and the tender coconut ice cream at natural’s 🤤 (should also try indo-chinese / momos in delhi - maybe at berco’s!) cc @mziizm :)

[aakanksha @____aakanksha]

🚨 Internship alert!! 🚨

Excellent opportunity to work at the precipice of modern day post-training! Model merging remains to be extremely effective but an enigma nevertheless. Come and figure out the mysteries behind model merging with Tom - he’s a delight to work with!

[aakanksha @____aakanksha]

@prachiruns Also horrified seeing the state of matters in Delhi around this - would love to get involved in making a difference here in case you’re hit up with any active initiatives / plans that you might be able to share!

[aakanksha @____aakanksha]

two years ago, i prepared and submitted my application on the very last day for this program and it changed my life. also saw a rainbow that day, so maybe my luck was shining.

don’t be me. apply today.

unless you’re certain you’ll see a rainbow on the 29th lol

(happy to help!)

[Cohere Labs @Cohere_Labs]

Applications are now open for the next cohort of the Cohere Labs Scholars Program! 🌟

This is your chance to collaborate with some of the brightest minds in AI & chart new courses in ML research. Let's change the spaces breakthroughs happen.

Apply by Aug 29.

[aakanksha @____aakanksha]

@ahmetustun89 @sarahookr Massive congrats, Ahmet! So lovely to see the photos! 💌🥰

[aakanksha @____aakanksha]

@viraataryabumi hard to tell if verb or noun 🤔 jk - congrats!

[Cohere Labs @Cohere_Labs]

Over 7000 languages are spoken worldwide 🌐, but AI safety efforts focus on only a fraction of them.

Our latest paper draws on our multi-year efforts with the wider research community to explore why this matters and how we can bridge the AI language gap.

[aakanksha @____aakanksha]

the complete cooking guide with all the ingredients, seasonings and garnishes for this soup of a model is here! 🍲🧂🌶️🔥

couldn’t be more proud to have continued my exploration of model merging and translated it to our A-class flagship model, Command A, with the best team! ✨

[Cohere @cohere]

We’re redefining what’s possible with AI.

With the release of our latest model, Command A, optimized for real-world agentic and multilingual tasks, we’re demonstrating our commitment to bringing enterprises AI that goes beyond the ordinary, and offers security & efficiency.

Our team has developed highly capable and efficient models that can be run on just 2 GPUs. Check out our tech report to learn more: https://t.co/4q814wC0G7

[aakanksha @____aakanksha]

A for A1 ✨🚀🔥

the scene just got a whole lot better with the arrival of this new powerhouse!

thrilled to have contributed towards its post-training!

[Cohere @cohere]

We’re excited to introduce our newest state-of-the-art model: Command A!

Command A provides enterprises maximum performance across agentic tasks with minimal compute requirements.

[aakanksha @____aakanksha]

Come check out our poster at the Safe Generative AI workshop in East Exhibition Hall A today!

Should be hard to miss - we’re manned right as you enter! 😛

#NeurIPS2024

[Marzieh Fadaee @mziizm]

Happening today!

[Cohere Labs @Cohere_Labs]

🔷 Check out “Mix Data or Merge Models? Optimizing for Performance and Safety in Multilingual Contexts” with @____aakanksha @aahmadian_ @beyzaermis @mziizm and @sarahookr

📜 https://t.co/3t5eJZkIFQ

[aakanksha @____aakanksha]

Started the morning off with the BEST pain au chocolat and views in Vancouver for #NeurIPS2024 🥐🏔️

Presenting this work at the Safe Generative AI workshop on Sunday, December 15!

I’ll also be at the Cohere booth later today to chat all things model merging, safety and more! 💫

[aakanksha @____aakanksha]

🚨 Our paper on model merging just dropped 🚨

Excited to share our work comparing merging 👷 vs data mixing 🧑‍🍳 as a recipe for building high-performant aligned LLMs!

Absolute pleasure cooking this up with the dream team @aahmadian_ @beyzaermis @seraphinagt @mziizm @sarahookr ✨

[Yannis Flet-Berliac @yfletberliac]

Come join us in a couple of minutes for our poster presentation of CoPG!

Paper: https://t.co/DREQyk9ToM

See you at Riverfront Hall @emnlpmeeting

@cohere @CohereForAI

[Cohere Labs @Cohere_Labs]

It's day 2 at @emnlpmeeting and we're excited to share research, connect with and learn from more researchers from around the world who have gathered in Miami.

Thanks to all those who checked out our lab's research yesterday, presented by @____aakanksha and @lushimabucoro.

[Sara Hooker @sarahookr]

Go chat to @____aakanksha as she presents our work today at #EMNLP2024 on expanding consideration of harms. 🔥

Very proud of this work with @aahmadian_ @beyzaermis @seraphinagt @KreutzerJulia @mziizm.

Tuesday, 12th November in Session 3 from 2-3:30pm @ Riverfront Hall ✨

[aakanksha @____aakanksha]

Also going to be at the Cohere booth on Wednesday and Thursday - come say hi! Happy to chat all things multilinguality, safety and more! 💫

[aakanksha @____aakanksha]

Arriving in Miami today for #EMNLP2024 to present our work on state-of-the-art alignment methods for mitigating “local” and “global” harms across a wide variety of languages! 🌴☀️🌊

Check out our poster on Tuesday, 12th November in Session 3 from 2-3:30pm @ Riverfront Hall! ✨

[aakanksha @____aakanksha]

Is multilingual safety a one-size-fits-all problem? Or are there nuances peculiar to cultures that must be addressed carefully?
Excited to share our state-of-the-art alignment efforts that address global and local harms across many languages! 🌍

📝 Paper: https://t.co/ykq4qhzexX

[Sara Hooker @sarahookr]

@fffffelipec lol we have :) but I completely agree -- plenty of fun work to be done ahead.

Shoutout to @____aakanksha for leading here with
@aahmadian_ @beyzaermis @seraphinagt @mziizm✨

https://t.co/b8gU1wNWxn

[aakanksha @____aakanksha]

@TingchenFu @CohereForAI @aahmadian_ @beyzaermis @seraphinagt @mziizm @sarahookr For more clarity, the data mixing approach is primarily based on the 15% Safety Mix, which implies that only 15% of all samples in that mix are safety-related (Section 2.2)

[aakanksha @____aakanksha]

@TingchenFu @CohereForAI @aahmadian_ @beyzaermis @seraphinagt @mziizm @sarahookr Hey Tingchen, glad to hear you thought the paper was exciting! While the absolute number of samples used for the separate data mixes (0%, 15% and 100% Safety) is different, the number of epochs they were trained for were adjusted such that all three saw the same amount of data.