Overview
Flow ID: select-embedding-model
Category: Model Management
Estimated Duration: 1-2 minutes
User Role: All Users
Complexity: Simple
Purpose: Choose which embedding model will be used for converting text into vectors for semantic search. This model is used when creating datasets, querying datasets during chat, and processing documents.
Related Flows
- Upload Embedding Model - Add models to select
- Upload New Dataset - Uses embedding model
- Create New Blockify Job - Requires embedding model
Prerequisites
Before starting, users must have:
- Application running
- At least one embedding model uploaded
- Access to Settings page
Step-by-Step Flow
Main Path (Happy Path)
Step 1: Navigate to Settings
- User Action: Click "Settings" in navigation
- System Response: Settings page loads
Step 2: Find Embedding Model Setting
- User Action: Look for embedding model configuration (may be in Chat Options, Admin Overrides, or dedicated Embeddings tab)
- System Response: Setting area displays
- UI Elements Visible:
- Embedding model selector or dropdown
- Currently selected model shown
- List of available embedding models
- Visual Cues: Dropdown or selector interface
Step 3: Open Embedding Model Dropdown
- User Action: Click embedding model dropdown
- System Response: List of available embedding models appears
- UI Elements Visible:
- Dropdown list showing:
- Model names (e.g., "Jina Embeddings", "BGE-Small")
- Model types (all show "Embeddings")
- Current selection marked
- May show model details (dimensions, size)
- Dropdown list showing:
Step 4: Select Different Model
- User Action: Click on desired embedding model
- System Response:
- Dropdown closes
- Selected model becomes active
- Model may begin loading
- UI Elements Visible:
- Selected model name in dropdown
- Loading indicator if model needs initialization
- Save button (if not auto-save)
Step 5: Model Loads (if needed)
- User Action: Wait for embedding model to initialize (typically 10-30 seconds)
- System Response:
- Model loads into background worker
- Progress indicated
- UI Elements Visible:
- Loading status
- "Initializing embedding model..."
Step 6: Selection Completes
- User Action: Verify selection successful
- System Response:
- Model ready
- Success indication
- Settings saved
- UI Elements Visible:
- Selected model shown
- Ready status indicator
- Can proceed with dataset operations
Final Step: Embedding Model Selected
- Success Indicator:
- New model active
- Ready for dataset operations
- Settings persisted
- System State Change:
- New embedding model active
- Future dataset creations use this model
- Dataset queries use this model
- Next Possible Actions:
- Create dataset using this model
- Upload or activate dataset
- Use dataset query in chat
Error States & Recovery
Error 1: No Embedding Models Available
Cause: None uploaded yet
User Experience:
- Dropdown empty or shows "No models"
- Cannot select
Recovery Steps:
- Upload embedding model first
- Return and select
Error 2: Model Loading Fails
Cause: Insufficient memory or file error
User Experience:
- Error during loading
- Model won't activate
Recovery Steps:
- Try different embedding model
- Check system resources
- Restart application
Version History
| Date | Version | Author | Changes |
|---|---|---|---|
| 2025-10-04 | 1.1 | Iternal Technologies | Initial documentation |
Notes
Embedding Model Purpose: Converts text to numerical vectors for semantic similarity search. Required for:
- Creating searchable datasets
- Querying datasets during chat
- Blockify job completion
Best Practices:
- Choose model before creating datasets
- Use same model for all related datasets
- Smaller models faster, larger may have better accuracy
- Keep model consistent once datasets created
Common User Questions:
- "Which embedding model is best?" - Jina or BGE models recommended for general use
- "Can I change models later?" - Yes, but doesn't affect existing datasets
- "Do I need this if I don't use datasets?" - No, only for dataset features
- "What happens if I change models?" - New datasets use new model; existing unchanged
Trigger
What initiates this flow:
- User manually initiates
Specific trigger: User needs to select or change embedding model, typically because:
- Setting up dataset features for first time
- Switching between different embedding models
- Just uploaded new embedding model
- Creating dataset that requires specific embedding model
User Intent Analysis
Primary Intent
Set which embedding model will be used for semantic search and dataset operations.
Secondary Intents
- Optimize search quality vs. speed
- Match embedding model to dataset requirements
- Test different embedding approaches