AI

LiquidAI/LFM3-1.7B · Hugging Face

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 27/10/2026, 23:47:25 UTC.

Briefings

LiquidAI released the LFM3-1.7B text model on October 27, 2026, featuring a tri-hybrid architecture optimized for edge devices. It supports 256K-token context, 128K vocabulary, and 24 layers of attention. The model is pre-trained on 22T tokens with curated mid-training data and is available in 4-bit precision requiring only 1 GB of memory. It is designed for local agents, tool calling, and domain-specific fine-tuning, though it is not recommended for complex multi-step reasoning due to its small size.

Key points

• LFM3-1.7B is a text-only language model with 1.73B parameters, 24 layers, and 128,000 vocabulary tokens.
• The model supports 256K-token context length and is designed for local agents, tool calling, and structured outputs.

Context

LFM3 models are part of LiquidAI's family of local agents and specialized applications, designed for edge devices. The LFM3-1.7B is a base text model that serves as a starting point for continued pre-training and post-training, with support for llama.cpp integration.

Limitations

Evidence is limited to the LFM3-1.7B text model; vision, audio, and decision models are described separately. The 22T pre-training data includes 3.5T tokens of curated mid-training data with agentic traces, but specific breakdowns of this data are not detailed. The model is not recommended for agentic coding or complex multi-step reasoning due to its small size.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 28/10/2026, 00:21:34 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeihfbdenlwsjvl3dxmnirhkhenmepo6mnbtmx6eqxlp6eysnwnrnf4