Liquid AI releases LFM3-1.7B, a 1.73B local agent model
Original title: LiquidAI/LFM3-1.7B — new model trending #30 on Hugging Face
The post-trained text model uses a 256K context, about 1 GB at 4-bit, and the LFM Open License v1.0.
Liquid AI published LFM3-1.7B, a post-trained text-only model for local agents and specialized apps, trending on Hugging Face. It has 1.73 billion parameters and a tri-hybrid stack of 24 layers: 13 gated short convolutions, 5 sliding-window attention layers, and 6 global attention layers. The checkpoint supports a 256K-token context, a 128,000-token vocabulary, and 16 listed languages. Liquid AI says 4-bit weights use about 1 GB and that the model is post-trained for harnesses such as Hermes Agent, Pi, and OpenClaw, with single-GPU fine-tuning. It is not recommended for agentic coding or complex multi-step reasoning.
- Architecture: 13 gated short convolutions, 5 sliding-window, 6 global attention.
- Vocabulary is 128,000 tokens; license is LFM Open License v1.0.
- About 1 GB of memory at 4-bit, aimed at phones and laptops.
- Retrieval or search is advised for knowledge-heavy tasks.
- The broader LFM3 family also lists vision, audio, omni, and decision models.
From the original article
9 min read at huggingface.cotext-generation · 6,311 downloads · 147 likes · license other
By LiquidAI · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.