What happened

LiquidAI released LFM3-1.7B, an edge-oriented text-generation model, on Hugging Face, where it currently ranks #30 on the trending models list with 155 likes and 6,311 recent downloads. The model is post-trained from LFM3-1.7B-Base and is positioned as a "text model for local agents and specialized applications," built on the company's new tri-hybrid architecture.

Key details

• The model has 1.73B parameters across 24 layers in a hybrid design: 13 gated short convolution layers, 5 sliding-window attention layers, and 6 global attention layers.
• Context window is 256K tokens (input plus output combined), with a 128,000-token vocabulary.
• It supports 16 languages, including English, Chinese, Arabic, French, German, Japanese, Korean, and Spanish.
• It was post-trained for agentic harnesses such as Hermes Agent, Pi, and OpenClaw, and supports tool calling and structured outputs.
• Edge friendliness is the core selling point: at 4-bit precision it takes about 1 GB of memory, so it "fits on any phone or laptop."
• Fine-tuning for specific domains, formats, and tools is supported via quick single-GPU adaptation.
• It uses the transformers library with safetensors format and is tagged custom_code; the license is LFM Open License v1.0 (lfm1.0), a non-standard license whose terms should be reviewed before commercial use.

Why it matters

The release illustrates a new approach for small models at the edge: a hybrid architecture (convolutions plus sliding-window and global attention) delivers a 256K context window at the 1.7B scale, while 4-bit quantization keeps memory at roughly 1 GB, making local agents on phones and laptops practical. The makers are explicit about limits: the model suits local agents, tool calling, structured outputs, and domain-specific fine-tuning; knowledge-heavy tasks should rely on retrieval or search, and due to its small size it is "not recommended for agentic coding or complex multi-step reasoning." For developers focused on on-device deployment and local agents, it is a notable option to evaluate, though the custom license terms warrant attention.

Who it affects

• Edge and mobile developers: the ~1 GB memory footprint allows direct deployment on phones or laptops.
• Local agent application developers: the model is post-trained for agentic harnesses and supports tool calling and structured outputs.
• Teams needing multilingual coverage: 16 languages, including Chinese and Arabic.
• Teams pursuing domain specialization: single-GPU fine-tuning is supported; however, coding agents and complex reasoning workloads call for larger models.

Hugging Face 热门模型第 30 名：LiquidAI/LFM3-1.7B，155 赞，近期下载 6311 次。 标签：transformers, safetensors, lfm3, text-generation, liquid, edge, conversational, custom_code, en, ar, zh, fr

README：

<img src="https://huggingface.co/LiquidAI/LFM3-1.7B/resolve/main/assets/liquid-ai-banner.webp" alt="Liquid AI" style="width: 100%; max-width: 100%; height: auto; display: block; margin: 0;" />

API
Try
Docs
Blog

LFM3-1.7B

LFM3-1.7B is a text model for local agents and specialized applications, based on our new tri-hybrid architecture.

• Agentic training: post-trained for agentic harnesses such as Hermes Agent, Pi, and OpenClaw, with a 256K-token context window.
• Designed for the edge: at 4-bit precision, it takes about 1 GB of memory, so it fits on any phone or laptop.
• Made to be specialized: adapt the model to your domains, formats, and tools via quick single-GPU fine-tuning.

Find more information about the LFM3 family in our release blog post.

Model details

The LFM3 family includes base, text, vision, audio, omni, and decision models.

Property Specification

Checkpoint Post-trained, text-only language model

Parameters 1.73B

Architecture 24 layers: 13 gated short convolutions, 5 sliding-window attention, 6 global attention

Context length 256K tokens (input + output)

Vocabulary 128,000 tokens

Languages English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, Polish

License LFM Open License v1.0

LFM3-1.7B is designed for local agents, tool calling, structured outputs, and domain-specific fine-tuning. Use retrieval or search for knowledge-heavy tasks. It is not recommended for agentic coding or complex multi-step reasoning due to its small size.

Inference

Choose a checkpoint variant for your inference framework and hardware.

Checkpoint Framework D

The original belongs to Hugging Face 热门模型 and its authors and is reproduced here for reading. To have it removed, write to support@aisa.one.