KI-Konzepte

Grundbegriffe und Mechanik moderner Sprachmodelle und KI-Workflows.

Ensemble / Multi-Modell-Orchestrierung Attention-Mechanismus Reasoning Effort Recall Sampling (LLM) Thinking Budget Automatic Speech Recognition (ASR) $/MTok (Kosten pro Million Tokens) Input-Token Output-Token Pfadabhängigkeit (LLM-Output) Whisper Agentic RAG KI-Agent Audio-Normalisierung AutoGPT Batch-API Benchmark (KI) Browser Use Chain-of-Thought Chunking Computer Use Constitutional AI Context Engineering Context Precision Context Recall Kontextfenster Contextual Retrieval Cosine Similarity dBFS, Headroom & Clipping DPO (Direct Preference Optimization) Embedding Evals Faithfulness Few-Shot Prompting Fine-Tuning Function Calling GGUF GraphRAG Guardrails Halluzination Hybrid Search HyDE Indirect Prompt Injection Inferenz Jailbreak Keyword-Glossar (STT) KV-Cache LLM LLM-as-a-Judge LoRA MCP Meta-Prompting PEFT (Parameter-Efficient Fine-Tuning) Prompt Caching Prompt Engineering Prompt Injection Prompt Leaking Prompt Template QLoRA Quantisierung Query Expansion RAG RAGAS Rate Limit (KI) ReAct (Prompting) Reranking RLHF Role Prompting Self-Consistency Self-Refine Semantic Search SFT (Supervised Fine-Tuning) Spotlighting Stop Sequences Structured Output / JSON Mode Synthetic Data System-Prompt Temperature Tier-Pricing Token Tokenizer Tool Call Top-p / Top-k TPM/RPM Tree of Thoughts Vektor-Datenbank VRAM Whisper initial_prompt XML-Tags im Prompt Adapter (PEFT) Batch-Inferenz Catastrophic Forgetting Continued Pretraining Instruction Tuning Knowledge Distillation Model Card ORPO (Odds Ratio Preference Optimization) Prefix Tuning Safetensors Streaming (LLM) Zero-Shot