LFM2.5-8B-A1B is an 8 billion parameter language model published by LiquidAI on the Hugging Face platform. It belongs to the LFM series of foundation models and is offered for download and local execution through the Hugging Face ecosystem.
The model includes a defined chat template in Jinja format along with a specific tokenizer configuration. This configuration sets the beginning of sequence token to "<|im_start|", the end of sequence token to "<|im_end|", a pad token of "<|pad|", and disables the default system prompt. The template supports conditional preservation of thinking steps and includes macros for formatting argument values and parsing multimodal content that can handle text strings, image placeholders, and JSON-serialized items.
It is delivered as a repository on Hugging Face, where users can access the model weights and associated files for integration with the Transformers library or compatible inference engines. The surrounding platform provides access to models, datasets, and community resources under an emphasis on open source and open science.
No pricing, licensing terms, or additional capabilities are stated in the repository metadata.
In the Foundation models & chat space, LFM2.5 8B A1B takes a focused approach. It focuses on providing a high-performance 8B parameter model with specialized architecture for local or hosted inference. LFM2.5 8B A1B is an open-source project aimed at developers. LFM2.5 8B A1B is open source under the Apache-2.0 license. It runs on the web, the command line, and API.
Liquid AI builds and maintains LFM2.5 8B A1B, and it first shipped in 2018. The project is developed in the open on GitHub with 162.8k stars and 745 commits in the last 90 days.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do