Exploring the world of Small Language Models
summary
Small language models run in the millions to a few billion parameters, can be distilled from larger models using knowledge distillation, shrunk by pruning and sped with 8-bit quantization, and are shown in examples like DistilBERT, TinyLlama and Granite 3 to be ~40% smaller, ~60% faster while retaining ~97% of capability. They run on laptops, phones and edge devices, cost and energy drop sharply, and work well for narrow domain tasks such as a medical support chatbot where privacy and low latency matter.
AI-written summary · about
chapters (11)
- 0:00Small Language Models intro
- 1:03SLM use cases
- 2:06What defines SLMs
- 3:11Model parameters
- 5:39Building small language models
- 7:46SLM vs LLM
- 10:17Advantages of small language models
- 11:20When to use SLMs
- 12:05Examples of small language models
- 12:55Limitations of small language models
- 14:24Q&A
related talks
18mSmolLM: The Rise of Smol Models
22mSpecializing Small Language Models with less data
40mSurviving the Agentic Hype with Small Language Models
44mHow to build your own Tiny Language Model from scratch
36mProduction-Ready AI Agents: From LLMs to Small Language Models
30mLarge Language Models and Prompt Engineering for Developers
11mWhen Less Is More: A Technical Overview of LLMs and the Strength of Smaller Models
41mThe Fantastic Small Language Models (SLMs) and When to Use Them