LFM2.5-1.2B-Thinking
Model Description
LFM2.5-1.2B-Thinking is a ~1.17B-parameter “thinking” (reasoning-tuned) language model from Liquid AI’s LFM2.5 family, designed for efficient deployment (including on-device/edge scenarios).
It supports long-context usage (up to 32,768 tokens) and is trained/tuned with a focus on instruction following and reasoning-oriented behavior.
Quickstart
Follow the deployment instruction on this page to run with one line of code:
https://sdk.nexa.ai/model/LFM2.5-1.2B-ThinkingFeatures
- Reasoning-oriented: tuned for stronger step-by-step problem solving vs. base variants.
- Conversational AI: context-aware dialogue using a chat template format.
- Tool / function calling: supports tool-use patterns for agentic workflows.
- Long context: supports up to 32K context length.
- Multilingual: supports multiple languages (including English and several major world languages).
Use Cases
- On-device assistants and private “local-first” chat experiences
- Tool-using agents (structured actions via function calls)
- Document Q&A and summarization (especially when paired with retrieval)
- Structured extraction and classification tasks
Inputs and Outputs
Input:
- Text prompts or conversation history, typically formatted using the model’s chat template.
Output:
- Generated text (answers, explanations, reasoning responses).
- Optional structured tool calls when prompted for tool-use behavior.
License
This repo is licensed under the Creative Commons Attribution–NonCommercial 4.0 (CC BY-NC 4.0) license, which allows use, sharing, and modification only for non-commercial purposes with proper attribution. All NPU-related models, runtimes, and code in this project are protected under this non-commercial license and cannot be used in any commercial or revenue-generating applications. Commercial licensing or enterprise usage requires a separate agreement. For inquiries, please contact dev@nexa.ai
References (other relevant links)
