Open Models: The Secret Weapon for Next-Generation Software
summary
Argues open weights plus deployment tools let teams customize and control LLMs: Gemma family includes Gemma 2 (2B, 9B, 27B, 70B) with instruction‑tuned, code, poly/vision, shield and data variants; OSI draft checklist has 17 components for open‑source AI. Uses Google Model Garden / Vertex AI and kasas for download, fine‑tuning and deployment; demo fine‑tunes a Yoda persona, deploys it to a Vertex endpoint (works, hard to jailbreak), and runs a quantized int4 TensorFlow Lite model with MediaPipe offline at ~1.3 GB.
AI-written summary · about
chapters (8)
- 0:00Introduction to open models
- 7:27What are open models
- 13:52How to start with open models
- 18:06Gemma open models overview
- 26:39Fine-tuning and demos
- 36:36Deploying on devices
- 44:26Future of open models
- 46:58Q&A
related talks
15mGemma, DeepMind's Family of Open Models
39mFlutter Project with Gemma: A State-of-the-Art Open Model
35mOpen-source AI: why it matters and how to get started
40mBuilding amazing Front-End Project with Gemma: A State-of-the-Art Open Model
25mState of Open Source AI Coding Models
42mOpen Source LLMs — Powering AI Accessibility and Innovation
24mAccelerating AI on Edge
44mBuilding amazing Front-End Project with Gemma: A State-of-the-Art Open Model