Really tiny. We released GGUF quants, so with 8-bit precision and 4k context the 350M will use ~380 MB and the 1.2B ~1.4 GB (to give you a rough estimate).
@maximelabonne
-
General-Purpose Language Models Excel at Summarization Tasks
By
–
We don't have a model specialized in summarization (yet), but the general-purpose LFMs are already very good at it!
-
All AI models now available on Hugging Face platform
By
–
All models are available today on @huggingface
! If you have a precise skill in mind for a task-specific model, please let me know in the comments! -
Model Release Enables Developers to Build AI Apps Locally
By
–
I'm happy it's released, it was such a cool post-training project Now I hope that app developers can take these models and build cool stuff with them, with less reliance on costly and slow cloud models. There's so much to do, life is too short to optimize ChatGPT prompts!
-
350M Math Reasoning Model Outperforms Larger Competitors
By
–
Finally, another personal favorite: a 350M math reasoning model. It turned out to be an absolute beast, outperforming Qwen3-0.6B and MobileLLM-R1. Note that it's also trained to output short CoTs, as a continuation of our work in this blog post: https://
liquid.ai/research/lfm-1
b-math-can-small-models-be-concise-reasoners
… -
LFM2-1.2B Achieves Parity with Larger Models for Edge
By
–
For latency purposes on edge devices, we wanted it to be a non-thinking model. That was a big challenge, but we managed to squeeze a ton of performance from LFM2-1.2B and perform on par with much bigger models on our internal bench (see figure) but also BFCL v3 and v4.
-
Tool Use Model Converts JSON Function Descriptions to Python Code
By
–
The tool use model is another cool one. It takes a traditional JSON description of functions as inputs (see the format) and outputs a list of Python functions It also supports 9 languages: English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, and Spanish.
-
RAG Model Excels at Long Context Processing and Information Retrieval
By
–
We also have a RAG model It's particularly good for processing long contexts and retrieving relevant information. It's very flexible and can adapt to various input styles.
-
350M Model Shows Power of Fine-Tuning for Big Data
By
–
It comes in two sizes: 1.2B and 350M I'm a big fan of the 350M model that can be used on GPUs to do big data operations. It's an absolute banger that shows how powerful fine-tuning can be
-
Data extraction LFM converts unstructured inputs to structured objects
By
–
Our data extraction LFM allows you to convert unstructured inputs into structured objects like JSON, XML, and YAML. You can specify the output language, but also the entire schema in the system prompt