LLM inference and fine-tuning on secure enterprise data
summary
Snowflake demonstrates concrete GenAI features: Document AI extracts values from PDFs with no‑code fine‑tuning; Snowflake Copilot generates SQL from natural language for real queries; Cortex adds vector storage and serverless LLM functions (Llama 2 available) and Streamlit hosts apps; Snowpark Container Services provides GPU_7 (A10G) pools for fine‑tuning and serving. A SentenceTransformer embedding model improves cosine similarity from ~0.55 to ~0.75 after fine‑tuning, gets registered in the model registry and deploys as a real‑time endpoint.
AI-written summary · about
chapters (10)
- 0:00Introduction to Snowflake and GenAI
- 3:50Snowflake platform overview
- 7:53Generative AI for enterprise data
- 9:37AI in seconds overview
- 12:19Snowflake Copilot demo
- 22:19Applications in minutes with Cortex
- 26:06Cortex specialized functions demo
- 33:32Customization in hours with Snowpark
- 37:12Fine-tuning embedding models demo
- 48:16Summary and Q&A
related talks
01h 19mSnowflake's AI research: A look inside our generative AI innovations
33mIntegrating Serverless LLMs with Data Platforms for Natural Language Analytics
29mBuilding End to End Gen AI Powered Analytics: From Data to Production
04mMaximize enterprise AI with seamless interoperability
15mEnterprise AI Runs on Trusted Data
17mPrepare to master common generative AI use cases with ease
36mImplementing Natural Language Analytics with Snowflake Cortex
30mBuilding Data AI Application with the Snowflake AI Data Cloud