Liquid AI

724 posts

Liquid AI banner
Liquid AI

Liquid AI

@liquidai

Build efficient general-purpose AI at every scale.

Cambridge, MA Katılım Mart 2023
47 Takip Edilen31.6K 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
511
3.9K
1.3M
Liquid AI retweetledi
sourcery
sourcery@sourceryy·
"Imagine it's not tokens that are actually important, it's just the outcome that actually matters." @liquidai CEO Ramin Hasani (@ramin_m_h) reacts to Palantir CEO Alex Karp's recent viral CNBC interview: " That was incredible. I fully agree with him." "I like that he said, 'The jig is up.'" "Enterprises now, actually in Europe, are waking up right now and they're getting into the place where they really see they have to take their time, similar to how they have been treating any other technology, to really understand the value of a tech and then deploy those solutions into their products." "In enterprise AI, as foundation model companies, we have to start thinking about that value chain a lot more. I think he was on the right path."
English
1
5
29
3.3K
Liquid AI
Liquid AI@liquidai·
We're celebrating an exciting milestone in our partnership with @Shopify : our LFMs have now processed 1 billion requests on Shopify’s platform! Read more about our multi-year partnership here: liquid.ai/blog/liquid-ai…
Liquid AI tweet media
English
8
14
130
33.7K
Liquid AI retweetledi
Molly O’Shea
Molly O’Shea@MollySOShea·
Liquid AI (@liquidai) CEO Ramin Hasani (@ramin_m_h) says: "We're bringing the cost of tokens to zero." "The axis was maximizing intelligence at all costs." Foundation models need to optimize across 3 axes: 1.) Intelligence and capability 2.) Efficiency and cost 3.) Substrate: where the intelligence actually runs "Efficiency is not an afterthought. Energy is not abundant." "If you maximize intelligence at all costs, that axis alone is not going to get you to the place that you want to go." "Efficiency and cost of intelligence as a first-class citizen, and not an afterthought." "Where does this intelligence system go? AI is majorly getting hosted in data centers, but you could also bring intelligence on phones, on laptops, on airplanes, on cars." "Imagine if it's not tokens that are actually important. It's just the outcome that actually matters." "Not just thinking about foundation models as the token machines that are generating money and revenue for the foundation model companies that are useless tokens, and getting them into the place where they can actually unlock true value for enterprises."
English
5
13
97
12.7K
Liquid AI retweetledi
Ramin
Ramin@ramin_m_h·
3-axises to optimize at the same time: 1) intelligence 2) efficiency 3) substrate masterful execution @RaiseSummit 👌🏻🇫🇷 and thanks @MollySOShea
Molly O’Shea@MollySOShea

Liquid AI (@liquidai) CEO Ramin Hasani (@ramin_m_h) says: "We're bringing the cost of tokens to zero." "The axis was maximizing intelligence at all costs." Foundation models need to optimize across 3 axes: 1.) Intelligence and capability 2.) Efficiency and cost 3.) Substrate: where the intelligence actually runs "Efficiency is not an afterthought. Energy is not abundant." "If you maximize intelligence at all costs, that axis alone is not going to get you to the place that you want to go." "Efficiency and cost of intelligence as a first-class citizen, and not an afterthought." "Where does this intelligence system go? AI is majorly getting hosted in data centers, but you could also bring intelligence on phones, on laptops, on airplanes, on cars." "Imagine if it's not tokens that are actually important. It's just the outcome that actually matters." "Not just thinking about foundation models as the token machines that are generating money and revenue for the foundation model companies that are useless tokens, and getting them into the place where they can actually unlock true value for enterprises."

English
1
2
25
2.2K
Liquid AI
Liquid AI@liquidai·
We are at @icmlconf 2026 in Seoul. Booth B116. We're hiring globally, including post-training and applied ML roles in our Tokyo office. Stop by our booth to talk! 私たちはICML 2026（ソウル）のブースB116に出展しています。 世界中で採用を行っており、東京オフィスではポストトレーニングおよび応用機械学習（Applied ML）のポジションも募集しています。ぜひブースにお立ち寄りいただき、お気軽にお話ししましょう！
Liquid AI tweet media
日本語
7
14
129
11.1K
Liquid AI retweetledi
Dave Blundin
Dave Blundin@DaveBlundin·
This is huge! If you've ever run small thinking models locally you'll recognize the doom-loop pattern. We welcome Antidoom with open arms! 🙌
Liquid AI@liquidai

Today we release Antidoom, an open-source method that removes a common failure mode in reasoning models: the doom loop. Doom-loop rates before and after, with eval scores up across the board: > Early LFM2.5-2.6B checkpoint: 10.2% → 1.4% > Qwen3.5-4B: 22.9% → 1% (greedy sampling) 🧵

English
4
6
96
11.1K
Liquid AI retweetledi
Justin Li 李泽天
Justin Li 李泽天@justinli9527·
Small models have much higher chance getting into doom loops than large model, now we present the solution to fix this😉
Liquid AI@liquidai

Today we release Antidoom, an open-source method that removes a common failure mode in reasoning models: the doom loop. Doom-loop rates before and after, with eval scores up across the board: > Early LFM2.5-2.6B checkpoint: 10.2% → 1.4% > Qwen3.5-4B: 22.9% → 1% (greedy sampling) 🧵

English
0
4
35
4.8K
Liquid AI retweetledi
Lior Alexander
Lior Alexander@LiorOnAI·
An open-source fix for one of the most common reasoning model failure modes. One of the biggest AI trends this year isn't larger models. It's systematically removing failure modes. Reasoning models sometimes get stuck repeating the same token sequence ("Wait...", "So...", "Alternatively...") until they exhaust the context window. Antidoom finds the single token that starts the loop and fine-tunes the model to prefer alternative next tokens at that position. It doesn't retrain the model from scratch, use RL, or teach the model new knowledge. On Qwen3.5-4B, doom loops dropped from 22.9% to 1%, and benchmark scores increased because the model stopped getting trapped before producing answers it was already capable of generating.
Liquid AI@liquidai

Today we release Antidoom, an open-source method that removes a common failure mode in reasoning models: the doom loop. Doom-loop rates before and after, with eval scores up across the board: > Early LFM2.5-2.6B checkpoint: 10.2% → 1.4% > Qwen3.5-4B: 22.9% → 1% (greedy sampling) 🧵

English
7
13
63
7.4K
Liquid AI
Liquid AI@liquidai·
RT @songdng: Don't let your models doom-loop 🔁 There's an antidoom now 💊
English
0
2
0
751
Liquid AI retweetledi
Leonie
Leonie@helloiamleonie·
Working with the @liquidai team on these engineering blogs is just so much fun! Here's what we've been working on: Reasoning models can get stuck mid-thought. That's what we call a "doom loop". A doom loop happens when the model repeats the same token ("Wait", "Let me reconsider") until the context window fills up. 3 things cause this: • 𝗢𝘃𝗲𝗿𝘁𝗿𝗮𝗶𝗻𝗲𝗱 𝘁𝗼𝗸𝗲𝗻𝘀: "Wait" and "Alternatively" dominate under uncertainty • 𝗦𝗲𝗹𝗳-𝗿𝗲𝗶𝗻𝗳𝗼𝗿𝗰𝗶𝗻𝗴 𝗰𝗼𝗻𝘁𝗲𝘅𝘁: each repeat raises the probability of the next one • 𝗚𝗿𝗲𝗲𝗱𝘆 𝘀𝗮𝗺𝗽𝗹𝗶𝗻𝗴: at low temperature, the most likely token always wins The standard fix is repetition_penalty at inference. It patches the symptom, not the root cause (and degrades performance). But the actual root cause is this: 𝗼𝗻𝗲 𝘁𝗼𝗸𝗲𝗻 𝘁𝗿𝗶𝗴𝗴𝗲𝗿𝘀 𝘁𝗵𝗲 𝗲𝗻𝘁𝗶𝗿𝗲 𝗹𝗼𝗼𝗽. So they built FTPO (Final Token Preference Optimization) to fix it at training time: 1. Identify the token that starts the loop mid-generation 2. Label it as rejected and sample up to 20 alternatives from the base model 3. Redistribute probability Doom-loop rate after training: LFM2.5-2.6B: 10.2% → 1.4% Qwen3.5-4B: 22.9% → 1% Code: github.com/Liquid4All/ant… Blog: liquid.ai/blog/antidoom
English
4
10
88
6.8K
Liquid AI retweetledi
Mathias Lechner
Mathias Lechner@mlech26l·
Small language models (<10B) often struggle with doom loops (=model stuck repeating itself) in long reasoning traces. We show that this can be fixed with specialized preference optimization
English
3
15
92
10.2K
Liquid AI retweetledi
Maxime Labonne
Maxime Labonne@maximelabonne·
New training technique to reduce doom loops! We applied it to LFM2.5-2.6B (SFT checkpoint) Qwen3.5-4B. By reducing doom loops, it also improves downstream evals. We open-source the training code and training dataset on @huggingface
Liquid AI@liquidai

Today we release Antidoom, an open-source method that removes a common failure mode in reasoning models: the doom loop. Doom-loop rates before and after, with eval scores up across the board: > Early LFM2.5-2.6B checkpoint: 10.2% → 1.4% > Qwen3.5-4B: 22.9% → 1% (greedy sampling) 🧵

English
22
48
496
48.5K
Liquid AI
Liquid AI@liquidai·
Runs in a few hours: for a 2-4B model, ~2 hrs to generate the training set on 8xH100, 1-2 hrs to train on a single H100. If your reasoning models loop on hard prompts, Antidoom recovers the accuracy those loops were costing you. Fully open source: > Blog: liquid.ai/blog/antidoom > Code: github.com/Liquid4All/ant… (6/6)
English
4
9
221
7.1K
Liquid AI
Liquid AI@liquidai·
A counterintuitive result: once looping is gone, near-greedy sampling wins. The belief that reasoning models need high temperature to explore may be conflating that with doom loops, which do most of their damage at low temp. Remove the loops and running hot stops helping, at least in the models we tested. (5/6)
Liquid AI tweet media
English
1
4
138
7.1K
Liquid AI
Liquid AI@liquidai·
Today we release Antidoom, an open-source method that removes a common failure mode in reasoning models: the doom loop. Doom-loop rates before and after, with eval scores up across the board: > Early LFM2.5-2.6B checkpoint: 10.2% → 1.4% > Qwen3.5-4B: 22.9% → 1% (greedy sampling) 🧵
GIF
English
85
283
2.3K
279.4K

Keşfet

@ramin_m_h @Shopify @RaiseSummit @MollySOShea @appvoidofficial @icmlconf @songdng @huggingface