BeamSearchOptimizer<T>
Models & Types · AiDotNet.PromptEngineering.Optimization
Optimizer that uses beam search to explore multiple promising prompt variations.
Inheritance: PromptOptimizerBase<T> → BeamSearchOptimizer<T>
For Beginners
Explores multiple paths simultaneously.
Example:
How it works:
- Keep track of top N prompts (the “beam”)
- Generate variations of all N prompts
- Score all variations
- Keep only the top N again
- Repeat
Benefits:
- More thorough than greedy search
- Less likely to get stuck in local optima
- Faster than exhaustive search
How It Works
Beam search maintains a fixed number of best candidates at each step, exploring variations of all candidates before selecting the top performers.
Constructors
| Constructor | Summary |
|---|---|
BeamSearchOptimizer(int beamWidth, int variationsPerCandidate, int? seed) |
Initializes a new instance of the BeamSearchOptimizer class. |
Methods
| Method | Summary |
|---|---|
void AddVariations(IEnumerable<string> prefixes, IEnumerable<string> suffixes, IEnumerable<string> insertions) |
Adds custom variations for beam search. |