Atlas · GenAI 2026
Edge AI
Small Language Models (SLM) & Edge AI
conceptPeak: 2025Efficient & Small ModelsAI consensus: 1/3
Prerequisites
- mediumTransformer Architecture
SLMs are compressed/distilled Transformers — understanding what is being compressed requires understanding the original architecture
- mediumModel Quantization
Edge deployment almost always requires quantization — these skills go hand in hand
Recommended reference
Abdin et al. (2024) 'Phi-3 Technical Report' — arXiv:2404.14219; Microsoft's case study in SLM design and capability
Notes from AI deep research
Anthropic Opus
Phi-3, Llama-Micro — <8B na telefonach. Quantization + distillation + SLM = edge stack
Google Deep Think
Ultra-lekkie <8B (Phi, Llama-Micro) [G#25]
Related skills
- → is subcategory of: AI(3/3)
- → is subcategory of: Machine Learning(3/3)
- ← is part of: Knowledge Distillation(2/3)
- ← is part of: Model Quantization(2/3)
- ← is an instance of: Ollama(2/3)
- → is subcategory of: Deep Learning(1/3)
- ← is an instance of: NVIDIA Jetson(0/3)
- ← is an instance of: LiteRT (TensorFlow Lite)(0/3)