Model: https://t.co/ma3Jpv7dzf Demo: https://t.co/Q41RxLc0eJ

LFM2-ColBERT-350M is also very fast! Its inference speed is on par with GTE-ModernColBERT-v1 (only 150M parameters) for query and document encoding across various batch sizes.

In a world built for humans, humanoid robots are designed to plug right into existing infrastructure. In a recent webinar, KraneShares CIO @ahern_brendan, Senior Investment Strategist Derek Yan, CFA, and Robostore Founder Teddy Haggerty explored why they believe this simple fact

Even more interestingly, LFM2-ColBERT-350M is an excellent cross-lingual retriever. This means that it is capable of retrieving documents based on queries from other languages.

We extended the NanoBEIR benchmark to include Japanese and Korean languages. We open-sourced this dataset on Hugging Face at LiquidAI/nanobeir-multilingual-extended for reproducibility. On this NanoBEIR benchmark, LFM2-ColBERT-350M displays significantly stronger multilingual

Late interaction retrievers like LFM2-ColBERT-350M are particularly interesting because they preserve much of the expressivity of re-rankers while retaining the efficiency of bi-encoders.

I went from: - XXL Shirts - 44 waist pants - 325 lbs - ~40% body fat To: - M/L shirts - 34 waist pants - 215 lbs - ~15% body fat In 1 year. You can do the same. Here's how in 5 easy steps👇

💍 LFM2-ColBERT-350M: One Model to Embed Them All Very happy to announce our first embedding model! It's a late interaction retriever with excellent multilingual performance. Available today on @huggingface!

Available today on @huggingface

LFM2-VL-3B just dropped! It's a bigger version of our VLMs with fast inference and strong performance. Look at how gracefully it distinguishes dogs from ice cream scoops

@maxencelsb @SinoueG Models: https://t.co/I4NTRSgZH6 Here's the paper

📚 Efficient Language Specialization for Small Language Models @maxencelsb and @SinoueG have released a preprint about their excellent work on fine-tuning small models in French. It shows a solid post-training pipeline to improve French performance while preserving English

@huggingface

New LFM2 release 🥳 It's a Japanese PII extractor with only 350M parameters. It's extremely fast and on par with GPT-5 (!) in terms of quality. Check it out, it's available today on @huggingface!

Real prop traders need the best—and now the #1 trading platform for prop is about to get even better. Follow Tradovate Prop for updates.

Apple wasn’t kidding, the iPhone 17 Pro is really built for running LLMs Here’s LFM2 8B A1B by @LiquidAI_ running on-device with MLX in @LocallyAIApp, the iPhone runs the 8B model with zero struggle Thanks @Prince_Canuma for the port to MLX, it made the MLX Swift port possible

Day 1 of the @LiquidAI_ fine-tuning hackathon in Tokyo this weekend. Jointly organized with @weights_biases and @LambdaAPI

@adrgrondin @LocallyAIApp @LiquidAI_ That is a very nice model for the iPhone. - MoE = very fast generation - Mostly conv layers = low memory footprint and fast for long context.

@huggingface LFM2-8B-A1B and GGUF quants are now available on Hugging Face! https://t.co/QxHyipGjnp

@huggingface Find more information about it and how we designed an MoE for CPU inference in the blog post https://t.co/VKY4L80Qvy

@huggingface We also have a nice preference optimization combining length-normalized DPO and APO. 👀

We just built and released the largest dataset for supervised fine-tuning of agentic LMs, 1.27M trajectories (~36B tokens)! Up until now, large-scale SFT for agents is rare - not for lack of data, but because of fragmentation across heterogeneous formats, tools, and interfaces.

@huggingface Compared to the 2.6B, this one has a big boost in knowledge capacity but also code!

@huggingface We have results for 16 different benchmarks This new model outperforms our LFM2-2.6B and similar-sized models. It's really good at math but also creative writing.