Skip to content

TESTING.md

Latest commit

 

History

History
206 lines (149 loc) · 5.73 KB

TESTING.md

File metadata and controls

206 lines (149 loc) · 5.73 KB

Sprint 3 Testing Guide - Streamlit UI

This guide covers testing the Streamlit UI for NeuralNav.

Prerequisites

  1. Ollama running with llama3.1:8b model
  2. FastAPI backend running on http://localhost:8000
  3. Streamlit and dependencies installed in virtual environment

Setup

# Activate virtual environment
cd backend
source venv/bin/activate

# Install UI dependencies (if not already installed)
pip install streamlit requests

Running the UI

Option 1: Using the startup script (Recommended)

# From project root
scripts/run_ui.sh

Option 2: Manual start

# From project root
cd backend
source venv/bin/activate
streamlit run ../ui/app.py

Note: On first run, Streamlit may prompt for an email address. Just press Enter to skip.

The UI will start on http://localhost:8501

Testing Checklist

1. UI Loads Successfully

  • Streamlit app starts without errors
  • Header displays "NeuralNav" with app icon
  • Sidebar shows app title and navigation
  • Two-column layout visible (Conversation | Recommendation)
  • Chat input field is present

2. Backend Connectivity

  • No connection errors in UI
  • Health check passes (check backend logs)

3. Example Prompts

Test each example prompt button in the sidebar:

  • Example 1: "Customer service chatbot for 5000 users, low latency critical"

    • Should recommend: Mistral 7B on A100-80GB
    • Cost: ~$3,285/month
    • SLO Status: ✅ MEETS SLO

  • Example 2: "Code generation assistant for 500 developers, quality over speed"

    • Should recommend: Mistral 7B on A10G
    • Cost: ~$730/month
    • SLO Status: ✅ MEETS SLO

  • Example 3: "Document summarization pipeline, high throughput, cost efficient"

    • Should recommend: Granite 8B on L4
    • Cost: ~$365/month
    • SLO Status: ✅ MEETS SLO

4. Chat Interface

  • User messages appear in chat
  • Assistant responses appear with recommendation summary
  • Spinner shows during API call
  • Messages persist in chat history

5. Recommendation Display

Test all tabs when a recommendation is displayed:

Overview Tab

  • SLO status badge displays correctly (✅ MEETS SLO or ⚠️ DOES NOT MEET SLO)
  • Model name and ID shown
  • GPU configuration displayed (count, type, tensor parallel, replicas)
  • Key metrics cards show: TTFT, TPOT, E2E, Throughput
  • Reasoning section displays

Specifications Tab

  • Use case and requirements shown
  • Traffic profile values displayed
  • SLO targets displayed
  • "✏️ Enable Editing" button works
  • Fields become editable when editing mode enabled
  • "💾 Save Changes" and "❌ Cancel" buttons work in edit mode

Performance Tab

  • TTFT metrics shown (p50, p90, p99)
  • TPOT metrics shown (p50, p90, p99)
  • E2E latency metrics shown (p50, p90, p99)
  • Throughput metrics shown (QPS, tokens/sec)
  • Delta values vs targets displayed

Cost Tab

  • Hourly cost displayed
  • Monthly cost displayed
  • GPU configuration details shown
  • Cost assumptions info box visible
  • "Generate Deployment YAML" button shows (Sprint 4 placeholder)
  • "Deploy to Kubernetes" button shows (Sprint 6 placeholder)

6. Navigation & UX

  • "🔄 New Conversation" button clears state and reloads
  • Switching between tabs works smoothly
  • Custom CSS styling applied (professional theme, metric cards)
  • Responsive layout works on different screen sizes

7. Error Handling

Test error scenarios:

  • Backend not running → Shows connection error message
  • Invalid prompt → Shows appropriate error
  • API timeout → Shows timeout error

Custom Test Scenarios

Try these additional prompts to test the system:

  1. High-volume scenario:

    • "I need to serve 10,000 concurrent users with a recommendation engine. Latency is very important - users expect results in under 300ms."
    • Should recommend higher-end GPU configuration

  2. Cost-sensitive scenario:

    • "Small team of 50 developers need a coding assistant. Budget is very limited, quality can be moderate."
    • Should recommend cost-effective configuration (L4 GPUs)

  3. Quality-focused scenario:

    • "Building a medical diagnosis assistant for 200 doctors. Accuracy is critical, budget is flexible, latency can be 1-2 seconds."
    • Should recommend larger model with better quality

Known Limitations (Sprint 3)

  • Editing specifications doesn't trigger re-recommendation (Sprint 4)
  • YAML generation not implemented (Sprint 4)
  • Kubernetes deployment not implemented (Sprint 6)
  • Conversation history not persisted across sessions
  • No authentication/user management

Success Criteria

Sprint 3 is successful if:

  1. ✅ UI loads without errors
  2. ✅ All 3 example scenarios work end-to-end
  3. ✅ Recommendation details display correctly in all tabs
  4. ✅ Edit mode toggles properly
  5. ✅ Chat interface is intuitive and functional
  6. ✅ No connection errors when backend is running

Troubleshooting

Streamlit won't start

# Check virtual environment
which python  # Should point to backend/venv

# Reinstall streamlit
pip install --upgrade streamlit

Backend connection fails

# Check FastAPI is running
curl http://localhost:8000/health

# Should return: {"status":"healthy","service":"ai-pre-deployment-assistant"}

Recommendation request fails

  1. Check backend logs for errors
  2. Verify Ollama is running: curl http://localhost:11434/api/tags
  3. Test backend directly: 
    curl -X POST http://localhost:8000/api/v1/recommend \
  -H "Content-Type: application/json" \
  -d '{"message": "test chatbot for 100 users"}'

UI shows old cached data

# Clear Streamlit cache
streamlit cache clear