Picture

Ramin Hasani @ramin_m_h

building @LiquidAI_ raminhasani.com New York, USA Joined July 2012
  • Tweets
    1K
  • Followers
    5K
  • Following
    441
  • Likes
    7K
Ramin Hasani retweeted
DeryaTR_ Profile Picture

Derya Unutmaz, MD @DeryaTR_

2 days ago

Wow! A 350-million-parameter AI model is performing as well as models that are 10× larger! @LiquidAI_ LFM2 is becoming a highly optimal model for edge devices, where size matters but smaller is better!

LiquidAI_ Profile Picture

Liquid AI @LiquidAI_

2 days ago

Wow! A 350-million-parameter AI model is performing as well as models that are 10× larger! @LiquidAI_ LFM2 is becoming a highly optimal model for edge devices, where size matters but smaller is better!

Ramin Hasani retweeted
romu_nishi Profile Picture

Rom Parnichkun @romu_nishi

2 days ago

Very proud of our team at Liquid AI Japan! We’ve just released our first Japanese task-specific SLM in the model library (leap.liquid.ai/models), with many more to come. It’s a small 350M model (i.e. reaches 200 tok/s prefill 40tok/s decode on a Raspi5), so you may notice a…

LiquidAI_ Profile Picture

Liquid AI @LiquidAI_

2 days ago

Very proud of our team at Liquid AI Japan! We’ve just released our first Japanese task-specific SLM in the model library (leap.liquid.ai/models), with many more to come. It’s a small 350M model (i.e. reaches 200 tok/s prefill 40tok/s decode on a Raspi5), so you may notice a… https://t.co/YpskK2c8Vb

ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

2 days ago

We got an LFM2 350M model to perform as good as GPT4o! > High quality general Japanese <> English translation > LFM2-350M outperforming models 10x larger > Notice the distance to Gemma 3 270M > Open weights in HF, and LEAP. The true realization of "ML done right", one of our…

LiquidAI_ Profile Picture

Liquid AI @LiquidAI_

2 days ago

We got an LFM2 350M model to perform as good as GPT4o! > High quality general Japanese <> English translation > LFM2-350M outperforming models 10x larger > Notice the distance to Gemma 3 270M > Open weights in HF, and LEAP. The true realization of "ML done right", one of our…

Ramin Hasani retweeted
LiquidAI_ Profile Picture

Liquid AI @LiquidAI_

2 days ago

Can we get a 350M parameter model to perform as good as GPT4o on specialized tasks? Today, we release an instance of our LFM2-350M, fine-tuned to perform competitively with GPT-4o on real-time general bi-directional Japanese <> English translation of short to medium context.…

Ramin Hasani retweeted
maximelabonne Profile Picture

Maxime Labonne @maximelabonne

2 days ago

Liquid AI Japan cooked with this 350M param model on par with GPT-4o for English ↔ Japanese translation That's a really nice example of fine-tuning done right 👌

Ramin Hasani retweeted
omkizzy Profile Picture

omkaar in SF @omkizzy

2 days ago

I wrote a short article about LFM-2's (by @LiquidAI_ ) hybrid architecture w/ illustration + simple pytorch impl.

Ramin Hasani retweeted
WesRothMoney Profile Picture

Wes Roth @WesRothMoney

3 days ago

Liquid AI has open-sourced two compact vision-language models, LFM2-VL-450M (350 M LM + 86 M vision) and LFM2-VL-1.6B (1.2 B LM + 400 M vision). Both deliver near-real-time inference that’s about 2× faster on GPUs than many existing VLMs while preserving competitive accuracy.…

ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

4 weeks ago

Liquid AI has open-sourced two compact vision-language models, LFM2-VL-450M (350 M LM + 86 M vision) and LFM2-VL-1.6B (1.2 B LM + 400 M vision). Both deliver near-real-time inference that’s about 2× faster on GPUs than many existing VLMs while preserving competitive accuracy.… https://t.co/ulHwi1yEUn

Ramin Hasani retweeted
jxmnop Profile Picture

Jack Morris @jxmnop

2 weeks ago

first i thought scaling laws originated in OpenAI (2020) then i thought they came from Baidu (2017) now i am enlightened: Scaling Laws were first explored at Bell Labs (1993)

Ramin Hasani retweeted
jackclarkSF Profile Picture

Jack Clark @jackclarkSF

6 days ago

Five years ago the frontier of LLM math/science capabilities was 3 digit multiplication for GPT-3. Now, frontier LLM math/science capabilities are evaluated through condensed matter physics questions. Anyone who thinks AI is slowing down is fatally miscalibrated.

ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

a week ago

I met Steve Pagliuca, the legendary ex-chairman of Bain Capital back in october 2023, thanks to John Werner. he grilled me for a couple of hours about who I am, what I do, and what I care about in life before getting into Liquid AI. I did the same! He then co-led our seed II…

ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

a week ago

Over 1 million Liquid foundation models downloaded through @huggingface! The community realized how far we can push with tiny models when they are designed from first principles. Proud of my team at @LiquidAI_! Liquid Discord community: discord.com/invite/liquid-… Play with our…

Ramin Hasani retweeted
LiquidAI_ Profile Picture

Liquid AI @LiquidAI_

a week ago

We closed out our first Liquid Hack 01 competition last week. Take a look at just a few of the private, mobile apps the community built. All powered by LFM2 and LEAP. Ready for the next competition? Details in 🧵

Ramin Hasani retweeted
DeryaTR_ Profile Picture

Derya Unutmaz, MD @DeryaTR_

a week ago

Big congrats to Ramin, @xanamini and rest of the @LiquidAI_ team! In my opinion, Liquid is the most innovative tiny AI foundation model, and I think the community is beginning to realize it. Still underrated, though!

ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

a week ago

Big congrats to Ramin, @xanamini and rest of the @LiquidAI_ team! In my opinion, Liquid is the most innovative tiny AI foundation model, and I think the community is beginning to realize it. Still underrated, though!

ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

a week ago

AMD is undervalued!

Ramin Hasani retweeted
victormustar Profile Picture

Victor M @victormustar

a week ago

Cool Hugging Face demo by @LiquidAI_ using their LFM2 hybrid models, running 100% in the browser using transformers.js and you can use with your own MCP tools 🚀

Ramin Hasani retweeted
LiquidAI_ Profile Picture

Liquid AI @LiquidAI_

a week ago

Meet LFM2 MCP, an in-browser tool calling MCP allowing for fast and local agentic workflows. Built by @KarnikShreyas, extending @xenovacom's WebGPU work.

Ramin Hasani retweeted
ramin_m_h Profile Picture

Ramin Hasani @ramin_m_h

2 weeks ago

Even when you are on a flight with no connection you can play with LFMs in our Apollo app! the vision language one is particularly a lot of fun to interact with 👇🏻

Ramin Hasani retweeted
naval Profile Picture

Naval @naval

2 weeks ago

Founders can delegate everything except recruiting, fundraising, strategy, and product vision.

Trends for United States

Iowa
25,3 B posts
Baylor
7.314 posts
#UFCParis
24,3 B posts
Karen
224 B posts
Duke
23,7 B posts
San Jose State
3.914 posts
Penn State
6.882 posts
Drew Allar
Syracuse
4.881 posts
NC State
2.911 posts
#HookEm
2.367 posts
UConn
3.415 posts
East Texas A&M
1.411 posts
Cheryl Richardson
10,3 B posts
Kevin Jennings
Angeli
1.960 posts
Chandler Morris
Davey Johnson
7.756 posts
Elizabeth Warren
49,6 B posts
SJSU
2.010 posts

You might like

NeurIPS workshop and digital community | 🌐 geometry, algebra, topology + 🤖 deep learning + 🧠 neuroscience | Join us on slack! https://t.co/Run9wPnrDB
Symmetry and Geometry...
@neur_reps
5K Followers
Building AI that makes autonomous decisions using world models, artificial curiosity, and temporal abstraction @GoogleDeepMind
Danijar Hafner
@danijarh
22K Followers
Member of the technical staff @ Anthropic. Most (in)famous for inventing diffusion models. AI + physics + neuroscience + dynamics.
Jascha Sohl-Dickstein
@jaschasd
24K Followers
Chair Prof. in AI, HKU; Visiting Prof. of EECS, UCB New book on Principles of Intelligence: https://t.co/leZlkURb7j
Yi Ma
@YiMaTweets
101K Followers
Research at @OpenAI. Previously at @GoogleAI on the Brain Team. Doing work on #ReinforcementLearning and #MachineLearning
Ofir Nachum
@ofirnachum
5K Followers
International Conference on Learning Representations #ICLR2026. SPC is @BharathHarihar3 and GC is @cvondrick
ICLR 2026
@iclr_conf
52K Followers
Chief AI Scientist @databricks via MosaicML.
Jonathan Frankle
@jefrankle
20K Followers
Post-trainologer at FAIR. Into codegen, RL, equivariance, generative models. Spent time at Qualcomm, Scyfer (acquired), UvA, Deepmind, OpenAI.
Taco Cohen
@TacoCohen
27K Followers
@Cohere's research lab and open science initiative that seeks to solve complex machine learning problems. Join us in exploring the unknown, together.
Cohere Labs
@Cohere_Labs
24K Followers
@Physical_int
Karol Hausman
@hausman_k
35K Followers
Assistant Prof @Cambridge_Uni, works on AI for the physical sciences.
Miles Cranmer
@MilesCranmer
13K Followers
Figuring out AI @allen_ai, open models, RLHF, fine-tuning, etc Contact via email. Writes @interconnectsai Wrote The RLHF Book Mountain runner
Nathan Lambert
@natolambert
56K Followers
Assistant Professor MIT @medialab @MITEECS @nlp_mit || PhD from CMU @mldcmu @LTIatCMU || Foundations of multisensory AI to enhance the human experience.
Paul Liang
@pliang279
8K Followers
Founder and Chief Scientist @ Emmi AI, Associate Prof @jkulinz. Former @MSFTResearch, @UvA_Amsterdam, @CERN, @tu_wien
Johannes Brandstetter
@jo_brandstetter
5K Followers
Research @AnthropicAI (Co-lead Discovery team) 💼 Past: Gemini @GoogleDeepMind (Co-led Blueshift team) 🧠 LLM Reasoning / AI Scientist 🎒Traveling & Backpacking
Behnam Neyshabur
@bneyshabur
29K Followers