Hyperparameters: Temperature, Top-P & Penalties

« Structured Outputs & JSON Schema Mode« Previous RAG & Vector Embeddings Architecture »Next »

Calibrate inference parameters for exact factual consistency (0.0) or creative brainstorms (0.8+).

Decoding Parameters Explained

Temperature: Divides logits before softmax. 0.0 always selects the highest-probability token (best for coding and extraction). Higher values flatten probabilities.

Top-P (Nucleus Sampling): Dynamically cuts off the tail of low-probability tokens, considering only the top cumulative probability percentile.

Frequency Penalty: Penalizes tokens based on how many times they have already appeared, preventing monotonous looping.

Google AdSense Slot (inContentBanner)

Set your Publisher ID in src/config/ads.ts to display live ads.

Test Yourself With Exercise

+15 XP

For deterministic code generation or JSON extraction, what temperature setting is recommended?

« Previous: Structured Outputs & JSON Schema Mode Next: RAG & Vector Embeddings Architecture »