We also provide an inference and a fine-tuning Colab notebooks. LFM2-VL-1.6B: https://
huggingface.co/LiquidAI/LFM2-
VL-1.6B
… LFM2-VL-450M: https://
huggingface.co/LiquidAI/LFM2-
VL-450M
…
@maximelabonne
-
LFM2-VL Models: 1.6B and 450M Vision-Language Released
By
–
-
Liquid Releases Fast 450M and 1.6B Parameter VLM Models
By
–
Liquid just released two 450M and 1.6B param VLMs! They're super fast and leverage SigLIP2 NaFlex encoders to handle native resolutions without distortion. Available today on @huggingface
! -
LFM2 Function Calling Demo with WebGPU
By
–
Brilliant demo of LFM2 with function calling running with WebGPU https://t.co/oGSi1YCaHd
— Maxime Labonne @ ICLR (@maximelabonne) 6 août 2025Brilliant demo of LFM2 with function calling running with WebGPU
-
Automated reasoning switches and artifact pollution in AI systems
By
–
Yes, the most naive way of doing this is an automated reasoning switch like old Claude or KAT-V1 (
https://
arxiv.org/abs/2507.08297). But artifacts from reasoning always pollute non-reasoning tasks. This is a bigger issue imo. -
OpenAI Updates System Prompt Architecture with Developer Tools
By
–
Not new, but the system prompt is now an automated list of relevant info (name, date, tools). The "developer" prompt is the old system prompt. https://
github.com/openai/harmony -
GPT-OSS Chat Template Special Tokens Analysis
By
–
Wow gpt-oss has a very interesting chat template Special tokens for "channel" (analysis, commentary, final) and "constrain". Still pretty ChatML-like!
-
Hunyuan 7B Limitations: Manual Hybrid Reasoning Performance Issues
By
–
Very grateful for Hunyuan 7B, but it makes me increasingly skeptical about manual hybrid reasoning. Is it something that you personally use? In my experience, it's not user-friendly, consistently underperforms larger models, and catastrophically reduces chat capabilities.
-
BERT CPU Model Powers Smart Home Interface with Function Calling
By
–
It's super fun to use a BERT-sized model that can run on a CPU and power a smart home interface. Also, we ranted a bit about function calling templates with a comparison between JSON and Python. Hope you enjoy it! 🙂
-
LFM2 Constrained Generation Smart Home Control Demo
By
–
LFM2 Outlines Constrained generation is amazing to squeeze more performance from tiny models. We just released a blog post with @remilouf with a smart home control demo. We used LFM2-350M to save notes, close blinds, and handle displays.
-
Non-Reasoning AI Model: Implications and Interesting Aspects
By
–
Oh no, it's a non-reasoning model. So it's even more interesting, but for the opposite reasons… Sorry about this mistake. 🙁