Mehrnaz Mofakhami

@mhrnz_m

Research Scholar @Cohere_Labs, Prev. researcher @Mila_Quebec, WiML @NeurIPSConf'24 Mentorship Chair, Prev. Visiting Researcher @ServiceNowRSRCH

Montréal, Québec

Joined June 2019

698 Following

1,278 Followers

Replies Reposts Media

Pinned
Mehrnaz Mofakhami
@mhrnz_m
Sep 10
Happy to present Tiny Aya L2-Thinker, a massively multilingual model at 3.35B scale with 32K context length trained for in-language reasoning in 45 languages. 🌎 We take a data-centric approach, combining multilingual reasoning, non-reasoning, and English reasoning data and
3
Mehrnaz Mofakhami
@mhrnz_m
Oct 15
Had a lot of fun giving this talk! Check out the recording for a deep dive into our recent paper on multilingual reasoning 👇
Cohere Labs
@Cohere_Labs
Oct 15
Ever thought about how AI adapts to different languages and cultures? 🌎💬 @mhrnz_m, Research Scholar from Cohere Labs unpacks her research on in-language reasoning, check out the full session recording: youtu.be/bE6A5hOnM_8
00:00
Mehrnaz Mofakhami
@mhrnz_m
Oct 6
Check out our blog post for what we learned building Tiny Aya L2-Thinker, a model that reasons in your language! 🌍👇 cohere.com/blog/building-…
Cohere Labs
@Cohere_Labs
Oct 6
Curious how to get models to reason in your native language, not just answer in it? The answer lies in data-mixing. 🔀 Following the release of our paper, we wrote a blog post to walk through the key ideas behind Tiny Aya L2-Thinker, our latest multilingual model that reasons
1
Mehrnaz Mofakhami
@mhrnz_m
Sep 23
How can we optimize multilingual reasoning for a wide range of languages without having training data for all? Join us next Monday, Sept 28th at 11 AM ET as I unpack our recent paper on how data mixing shapes multilingual reasoning, followed by a live Q&A! Register here:
Cohere Labs
@Cohere_Labs
Sep 16
Join us on Monday, September 28 as @mhrnz_m, Research Scholar at Cohere Labs, discusses her latest paper, “Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning.” (arxiv.org/abs/2609.10445) Mehrnaz will share insights from the
Mehrnaz Mofakhami
@mhrnz_m
Sep 15
Check out our paper on @askalphaxiv with a nice summary and a 5-min podcast! alphaxiv.org/abs/2609.10445 Tiny Aya L2-Thinker beats models 2-7x its size on in-language reasoning rate and task accuracy on data-scarce languages.
Mehrnaz Mofakhami
@mhrnz_m
Sep 10
Happy to present Tiny Aya L2-Thinker, a massively multilingual model at 3.35B scale with 32K context length trained for in-language reasoning in 45 languages. 🌎 We take a data-centric approach, combining multilingual reasoning, non-reasoning, and English reasoning data and
2