yasserrmd/GLM4.7-Distill-LFM2.5-1.2B
yasserrmd/GLM4.7-Distill-LFM2.5-1.2B is a 1.2 billion parameter instruction-following language model, distilled from GLM-4.7 into the Liquid AI LFM2 architecture. Optimized for conciseness and strong instruction adherence, it focuses on delivering final-answer quality rather than verbose explanations. This model is efficient for local and edge deployments, making it suitable for assistant, agentic, and system integration use cases requiring clear, grounded responses.
Loading preview...
Model Overview
GLM4.7-Distill-LFM2.5-1.2B is a compact 1.2 billion parameter instruction-following language model developed by yasserrmd. It was created through offline distillation from the larger GLM-4.7 teacher model into the efficient Liquid AI LFM2 architecture. The primary design goals for this model include conciseness, strong instruction following, and efficiency for local and edge deployments.
Key Characteristics
- Architecture: Based on Liquid AI LFM2.
- Training Method: Offline supervised distillation (SFT with LoRA), using GLM-4.7 to generate instruction-response pairs.
- Behavior: Optimized for final-answer quality and clear, grounded, instruction-aligned responses, explicitly avoiding chain-of-thought reasoning.
- Efficiency: Designed for efficient operation in assistant, agentic, and system-integration scenarios.
Intended Use Cases
This model is well-suited for applications such as:
- General-purpose assistants and planning tasks.
- Summarization, explanation, and lightweight coding assistance.
- Agentic workflows and system integration/automation.
- On-device or edge inference where resource efficiency is critical.
Limitations
As a compact distilled model, it may exhibit limitations such as hallucination with insufficient premises, struggles with adversarial logical inference, and a lack of temporal awareness. For critical reasoning, external verification is recommended.