Liquid AI

768 posts

Liquid AI banner
Liquid AI

Liquid AI

@liquidai

Build efficient general-purpose AI at every scale.

Cambridge, MA Katılım Mart 2023
52 Takip Edilen32.1K Takipçiler
Sabitlenmiş Tweet
Liquid AI
Liquid AI@liquidai·
Today, we're releasing LFM2.5-8B-A1B, a device-optimized model designed to power real-life applications on phones, laptops, PCs, robots, and fast & lightweight server-side use-cases. > 8B MoE, 1.5B active > Expanded 128K context > LFM2.5 flagship hybrid MoE architecture > Trained on 38T tokens + large-scale RL > fast, reliable tool calling, punching above its weight, comparable to models with up to 4x its size > customizable on a single GPU for any specialized task > LFM2 open-weight license 🧵
Liquid AI tweet media
English
142
510
3.9K
1.3M
Liquid AI
Liquid AI@liquidai·
Multi-label classification. One forward pass. Zero completion tokens. We fine-tuned LFM2.5-Encoder-230M and LFM2.5-Encoder-350M to score every label at once: no decoding loop, nothing to parse or repair. First in a new encoder demo series, watch it in action. Check out the cookbook: github.com/Liquid4All/coo…
English
6
30
285
20.2K
Liquid AI retweetledi
Viviana Márquez
Viviana Márquez@vivmarquez·
Not every AI problem is a generation problem. For routing, classification, retrieval, policy checks, and similar tasks, generating an answer token by token can add unnecessary latency, cost, and complexity. A fine-tuned encoder can handle many of these tasks directly: one forward pass, structured outputs, predictable latency, and no free-form response to parse. That's why I'm so excited about the new LFM2.5-Encoders.
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
6
14
95
7.8K
Liquid AI retweetledi
Pau Labarta Bajo
Pau Labarta Bajo@paulabartabajo_·
I am not sure about the net impact of the mega large language models we hve today (aka too costly for you/the planet for the value each output token brings) As for Small models (like this) the net effect is clearly positive. Congratulations to my dear team at Liquid ❤️
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
3
10
68
6.8K
Liquid AI retweetledi
Jimmy Smith
Jimmy Smith@jimmysmith1919·
Check out our new encoder models!
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
0
2
21
1.4K
Liquid AI
Liquid AI@liquidai·
@DeryaTR_ many thanks for your continued support Derya 🙏🏻
English
0
0
5
204
Liquid AI retweetledi
Derya Unutmaz, MD
Derya Unutmaz, MD@DeryaTR_·
Liquid AI LFM models are fantastic for their size! People may not realize they are also amazing in biomedical research; I have firsthand experience with this. The founders are also super passionate about both open source and advancing biology. Grateful that I know them!
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
5
8
81
10.6K
Liquid AI retweetledi
tomaarsen
tomaarsen@tomaarsen·
Let's gooo, @liquidai released some base Encoder models, i.e. bidirectional models that should act as very solid base models to finetune for tasks like retrieval, PII detectors, classifiers, clustering, routers, etc. They're faster than ModernBERT at high seq. lengths too! 🧵
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
1
8
79
4.8K
Liquid AI retweetledi
Leonie
Leonie@helloiamleonie·
we have encoders now! you know what this means, right? 👀
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
21
13
335
49.7K
Liquid AI retweetledi
David Hendrickson
David Hendrickson@TeksEdge·
💡Do you need a small, fast, prompt routing tool? Then this free open-source encoder is for you. 🚀 Liquid AI released LFM2.5-Encoder 📊 230M & 350M bidirectional encoders ⚡ 3.7× faster than ModernBERT-base on CPU at 8K tokens 📈 350M ranks #4 on GLUE + SuperGLUE + multilingual tasks 🧠 Perfect for: • Prompt / intent routing • Classification • Long-doc triage before big models ✅ Open weights • Runs on CPU • 8K context No need to retrain for new routes. Just load it, attach a simple classification head, and you have a cheap prompt/intent router that runs great on CPU. 3.7× faster than ModernBERT at 8K tokens.
David Hendrickson tweet media
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
0
3
23
3.2K
Liquid AI retweetledi
Edoardo Mosca
Edoardo Mosca@EdoardMosca·
The bi-directional LFM family just got an upgrade!
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
0
1
12
851
Liquid AI retweetledi
Song
Song@songdng·
Tiny encoders that run blazingly fast on CPUs. Try the demos!
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
3
2
47
4.1K
Liquid AI retweetledi
Fernando Fernandes Neto
Fernando Fernandes Neto@FernandoNetoAi·
Agent systems are often discussed as if the central LLM does all the work. In practice, the surrounding harness also needs efficient models for routing requests, enforcing policies, extracting entities, and checking whether retrieved context is relevant. 🧵
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
2
7
38
31.1K
Liquid AI retweetledi
Ramin
Ramin@ramin_m_h·
tiny encoders for a new way of routing agents! enjoy ✨
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
1
3
23
1.5K
Liquid AI retweetledi
Maxime Labonne
Maxime Labonne@maximelabonne·
We just released two new encoder models (MLM) in 2026 🚀🚀🚀 They're super fast, easy to train, and strongly multilingual. Try them today: we created 5 demos on @huggingface
Liquid AI@liquidai

Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵

English
3
14
116
8.1K
Liquid AI
Liquid AI@liquidai·
Today we release LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: bidirectional encoders that stay fast at long context, even on CPU. > LFM2.5-Encoder-230M: about 3.7x faster than ModernBERT-base on CPU at 8,192 tokens. Under 30s per forward pass, versus over a minute and a half. > LFM2.5-Encoder-350M: 4th of 14 models on GLUE, SuperGLUE, and multilingual classification, behind only three larger models, one of them nearly 10x its size. 🧵
English
34
110
911
189K

Keşfet

@KLee4422 @paulabartabajo_ @DeryaTR_ @huggingface @elonmusk @BarackObama @taylorswift13 @cristiano