Skip to main content
Version: v2.10.0

LLM Model Selection

Overview​

Yeedu allows you to select different Large Language Models (LLMs) on a per-cell basis in notebooks. Choose the best model for each cell's task to optimize performance and cost.

Available Models​

Model Options​

ModelBest ForSpeedCostContext
GPT-4Complex tasks, reasoningMediumHigh128K tokens
GPT-3.5Quick suggestions, summarizationFastLow16K tokens
ClaudeCode analysis, documentationMediumMedium100K tokens
PaLMCreative tasks, translationMediumMedium30K tokens
Local ModelPrivacy, cost-freeFastFreeLimited

Model Capabilities​

GPT-4

  • Advanced reasoning
  • Complex problem-solving
  • Code generation and debugging
  • Multimodal input (text + images)

GPT-3.5

  • Fast inference
  • Good for simple tasks
  • Lower token usage
  • Cost-effective

Claude

  • Excellent code understanding
  • Long context windows
  • Good for documentation
  • Constitutional AI alignment

Selecting Model Per Cell​

Change Cell Model​

  1. Open notebook cell
  2. Click Model Selector dropdown
  3. Choose model from list
  4. Selection saved for this cell

Default Model​

Set default model for all new cells:

  1. Go to Notebook Settings
  2. Find AI Assistant section
  3. Select Default LLM Model
  4. New cells use this model
  5. Can override per cell

Model Indicator​

  • Cell shows current model in small label
  • Indicator color indicates model tier
  • Gold: Premium (GPT-4)
  • Blue: Standard (GPT-3.5)
  • Green: Alternative (Claude, others)

Using AssistantX with Model Selection​

Single Model Task​

  1. Select cell
  2. Choose specific model
  3. Write prompt for AssistantX
  4. Model processes your request
  5. Review generated response

Multi-Model Comparison​

Compare models for same task:

  1. Create 3 cells with same code request
  2. Assign different models to each cell
  3. Run all cells
  4. Compare results side-by-side
  5. Choose best output

Cost Considerations​

Model Pricing​

  • GPT-4: Higher cost (~$0.03 per 1K tokens)
  • GPT-3.5: Lower cost (~$0.0005 per 1K tokens)
  • Claude: Medium cost (~$0.008 per 1K tokens)
  • Local Model: No API cost (free)

Optimization Strategy​

  • Use GPT-3.5 for simple tasks
  • Use GPT-4 only when needed
  • Use Local Model for frequent, simple requests
  • Monitor token usage

Model Selection Guide​

When to Use GPT-4​

  • Complex code generation
  • Debugging intricate problems
  • Multi-step reasoning required
  • First-time complex tasks

When to Use GPT-3.5​

  • Quick suggestions
  • Simple code completion
  • Error messages explanation
  • Learning resources

When to Use Claude​

  • Code review and refactoring
  • Documentation writing
  • Deep code understanding
  • Technical explanation

When to Use Local Model​

  • Frequent, simple operations
  • Offline operation needed
  • Privacy-sensitive tasks
  • Budget constraints

Advanced Configuration​

Token Usage Tracking​

View token usage per model:

  1. Click cell with AI output
  2. See Tokens Used stat
  3. Track across cells
  4. Monitor total monthly usage

Model Parameters (Advanced)​

For each model, configure:

  • Temperature: Creativity (0=deterministic, 1=creative)
  • Max Tokens: Response length limit
  • Top-P: Diversity control
  • Frequency Penalty: Repetition reduction

Default settings work for most cases.

Model Context Window​

What is Context Window?​

Maximum tokens the model can read:

  • GPT-4: 128K tokens (~100 pages)
  • GPT-3.5: 16K tokens (~12 pages)
  • Claude: 100K tokens (~75 pages)

Using Large Context​

For large codebases:

  1. Select Claude (100K context)
  2. Include full codebase context
  3. Model understands full scope
  4. Provides better suggestions

Switching Models Mid-Workflow​

Change Model for Task​

If current model isn't working well:

  1. Select new model for cell
  2. Re-run with different model
  3. Compare outputs
  4. Keep the better result

Model Limitations​

Know the Boundaries​

  • Models can make mistakes
  • Always verify generated code
  • Context window limits large files
  • Token limits affect response length
  • API availability may vary

When Models Struggle​

If getting poor results:

  • Try different model
  • Provide more context
  • Simplify the request
  • Break into smaller tasks

Best Practices​

Model Selection Strategy​

✓ Do:

  • Match model to task complexity
  • Use cheaper models when possible
  • Test different models for comparison
  • Monitor total token usage
  • Consider latency for notebooks

✗ Don't:

  • Always use premium models
  • Ignore cost implications
  • Use very large models for simple tasks
  • Forget to verify output
  • Exceed API rate limits

Workflow Optimization​

  1. Start with GPT-3.5
  2. Upgrade to GPT-4 if needed
  3. Compare models side-by-side
  4. Save best-working models for templates
  5. Document model choices for repeatability

Troubleshooting​

Model Not Available​

  • Confirm model subscription
  • Check API credentials configured
  • Verify region availability
  • Try different model

Poor Quality Output​

  • Provide more context
  • Try different model
  • Rephrase the request
  • Break into smaller tasks

Slow Response​

  • Reduce context size
  • Use faster model (GPT-3.5)
  • Check API status
  • Monitor network latency

See Also​