Surviving the Agentic Hype with Small Language Models
summary
Claims that compact SLMs can do planning, tool calling, code glue and instruction following, with concrete demos: XLM2 32B quantized runs on a laptop (<16 GB RAM), a local Llama.cpp inference server handles prompts, and LangChain orchestrates tools; a data agent fetches Yahoo Finance prices, caches them in an Apache Iceberg lakehouse, and runs locally then deploys to cloud via Tower. Benchmarks show 111 models on the function‑calling leaderboard and a costly GLM at 355B, so quantized SLMs give practical, cheaper experimentation.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 3:11From LLMs to agents
- 9:00Why SLMs matter
- 17:55SLM agentic workflow setup
- 27:07Live demo: data agent with SLM
- 32:50Future of agentic workflows
- 34:37Q&A
related talks
36mProduction-Ready AI Agents: From LLMs to Small Language Models
41mThe Fantastic Small Language Models (SLMs) and When to Use Them
47mOwn Your LLM: Running Generative and Agentic AI Locally
11mWhen Less Is More: A Technical Overview of LLMs and the Strength of Smaller Models
27mBuilding Useful Apps With Small, Local LLMs
27mAI Demystified Unveiling the Intelligence Behind the Machine
30mLLMs: Beyond the Hype - A Practical Journey to Scale
20mSelf-hosting AI LLMs - a beginners guide