Video Walkthrough
The Azure ML Prompt Flow labs takes about an hour to complete end-to-end. It can take longer if you factor in time for exploration and data entry. You might also run into issues or have questions about what should have happened at a specific point in the lab, to help you debug or understand concepts better.
This video recording of the labs can help. The tables below show you the timestamps for each relevant step, within this video. Just visit the YouTube page and open the description to get pre-linked key moments to jump directly to the relevant segments to review walkthrough steps for reference.
1. Prepare Environment
- 00:03:30 | Launch GitHub codespace for Lab
- 00:04:31 | Prompt Flow overview (PowerPoint)
- 00:36:44 | Open VS Code. Login to Azure.
- 00:39:37 | Create Azure resource Set env variables. Save config to conflig.json
- 00:50:18 | Download conflig.json. Select Kernel. Run All.
2. Load Data into Vector Datastore
- 00:52:51 | Authenticate to Azure ML Studio
- 00:53:32 | load your data source and destination
- 00:55:20 | create compute instance
- 00:55:25 | create OpenAI connection
- 00:55:55 | use text embedding model deployment. submit pipeline job
- 00:56:47 | ⏳ Takes ~10 mins. Monitor Vector Index job.
- 00:57:12 | Get vector datastore path.
3. API connections and Compute runtime
4. Create a Chat Agent
- 01:09:45 | Add Embeddings for input
- 01:12:08 | Add Vector db Lookup. Use your own data
- 01:14:17 | Add System Prompt and rules
- 01:18:01 | Add response field
- 01:20:10 | Test the chat agent
- 01:24:33 | Add prompt variants
- 01:29:12 | Test Chat with another prompt variant