| Transformer | Architecture |
| Attention | Architecture |
| Self-attention | Architecture |
| Multi-head attention | Architecture |
| Feed-forward network | Architecture |
| Layer normalization | Architecture |
| Residual connection | Architecture |
| Positional encoding | Architecture |
| Rotary positional embedding | Architecture |
| Mixture of experts | Architecture |
| Sparse mixture of experts | Architecture |
| Dense model | Architecture |
| Decoder-only model | Architecture |
| Encoder-decoder model | Architecture |
| Embedding layer | Architecture |
| Output head | Architecture |
| Vocabulary | Architecture |
| Hidden state | Architecture |
| Activation function | Architecture |
| GELU | Architecture |
| SwiGLU | Architecture |
| Softmax | Architecture |
| Logits | Architecture |
| Autoregressive model | Architecture |
| Context window | Architecture |
| Attention mask | Architecture |
| Causal mask | Architecture |
| Grouped-query attention | Architecture |
| Multi-query attention | Architecture |
| Flash attention | Architecture |
| Sliding window attention | Architecture |
| State space model | Architecture |
| Recurrent layer | Architecture |
| Convolutional layer | Architecture |
| Normalization layer | Architecture |
| Dropout | Architecture |
| Weight tying | Architecture |
| Parameter count | Architecture |
| Model size | Architecture |
| Base model | Architecture |
| Pre-training | Training |
| Fine-tuning | Training |
| Instruction tuning | Training |
| Supervised fine-tuning | Training |
| RLHF | Training |
| DPO | Training |
| PPO | Training |
| GRPO | Training |
| Reward model | Training |
| Preference data | Training |
| Synthetic data | Training |
| Data curation | Training |
| Data cleaning | Training |
| Deduplication | Training |
| Tokenization | Training |
| Byte-pair encoding | Training |
| SentencePiece | Training |
| Learning rate | Training |
| Batch size | Training |
| Gradient accumulation | Training |
| Gradient checkpointing | Training |
| Mixed precision | Training |
| LoRA | Training |
| QLoRA | Training |
| Adapter | Training |
| Prefix tuning | Training |
| Prompt tuning | Training |
| Distillation | Training |
| Quantization | Training |
| Pruning | Training |
| Knowledge distillation | Training |
| Continual learning | Training |
| Catastrophic forgetting | Training |
| Curriculum learning | Training |
| Reinforcement learning | Training |
| Chain-of-thought training | Training |
| Self-play | Training |
| Constitutional AI | Training |
| Red teaming | Training |
| Data mixture | Training |
| Inference | Inference |
| Token | Inference |
| Prompt | Inference |
| Completion | Inference |
| Temperature | Inference |
| Top-p | Inference |
| Top-k | Inference |
| Beam search | Inference |
| Greedy decoding | Inference |
| Sampling | Inference |
| KV cache | Inference |
| Paged attention | Inference |
| Speculative decoding | Inference |
| Batching | Inference |
| Continuous batching | Inference |
| Throughput | Inference |
| Latency | Inference |
| Time to first token | Inference |
| Tokens per second | Inference |
| Streaming | Inference |
| Stop sequence | Inference |
| Logit bias | Inference |
| Repetition penalty | Inference |
| Frequency penalty | Inference |
| Presence penalty | Inference |
| Context length | Inference |
| Max output tokens | Inference |
| Truncation | Inference |
| Chunking | Inference |
| Prompt caching | Inference |
| Prefix caching | Inference |
| Quantized inference | Inference |
| FP8 inference | Inference |
| INT4 inference | Inference |
| GPU memory | Inference |
| VRAM | Inference |
| Model parallelism | Inference |
| Tensor parallelism | Inference |
| Pipeline parallelism | Inference |
| Expert parallelism | Inference |
| Benchmark | Evaluation |
| Leaderboard | Evaluation |
| Evaluation harness | Evaluation |
| Held-out set | Evaluation |
| Test set | Evaluation |
| Validation set | Evaluation |
| Contamination | Evaluation |
| Data leakage | Evaluation |
| Few-shot evaluation | Evaluation |
| Zero-shot evaluation | Evaluation |
| Chain-of-thought evaluation | Evaluation |
| Pass at k | Evaluation |
| Exact match | Evaluation |
| F1 score | Evaluation |
| BLEU | Evaluation |
| ROUGE | Evaluation |
| Perplexity | Evaluation |
| Calibration | Evaluation |
| Arena Elo | Evaluation |
| Pairwise preference | Evaluation |
| Human evaluation | Evaluation |
| LLM-as-a-judge | Evaluation |
| Rubric | Evaluation |
| Regression test | Evaluation |
| Eval suite | Evaluation |
| Task suite | Evaluation |
| Long-context evaluation | Evaluation |
| Multimodal evaluation | Evaluation |
| Safety evaluation | Evaluation |
| Jailbreak evaluation | Evaluation |
| Robustness evaluation | Evaluation |
| Factuality evaluation | Evaluation |
| Hallucination rate | Evaluation |
| Abstention rate | Evaluation |
| Coverage | Evaluation |
| Variance | Evaluation |
| Confidence interval | Evaluation |
| Statistical significance | Evaluation |
| Reproducibility | Evaluation |
| Benchmark saturation | Evaluation |
| Retrieval-augmented generation | Retrieval and RAG |
| RAG | Retrieval and RAG |
| Embedding | Retrieval and RAG |
| Vector database | Retrieval and RAG |
| Semantic search | Retrieval and RAG |
| Hybrid search | Retrieval and RAG |
| BM25 | Retrieval and RAG |
| Reranker | Retrieval and RAG |
| Chunking strategy | Retrieval and RAG |
| Document loader | Retrieval and RAG |
| Indexing | Retrieval and RAG |
| Recall | Retrieval and RAG |
| Precision | Retrieval and RAG |
| MRR | Retrieval and RAG |
| NDCG | Retrieval and RAG |
| Context injection | Retrieval and RAG |
| Citation | Retrieval and RAG |
| Grounded answer | Retrieval and RAG |
| Knowledge base | Retrieval and RAG |
| Document store | Retrieval and RAG |
| Metadata filter | Retrieval and RAG |
| Query rewriting | Retrieval and RAG |
| HyDE | Retrieval and RAG |
| Parent-document retrieval | Retrieval and RAG |
| Sentence-window retrieval | Retrieval and RAG |
| Multi-vector retrieval | Retrieval and RAG |
| Graph RAG | Retrieval and RAG |
| Agentic RAG | Retrieval and RAG |
| Evaluation set | Retrieval and RAG |
| Retrieval latency | Retrieval and RAG |
| AI agent | Agents |
| Tool use | Agents |
| Function calling | Agents |
| ReAct | Agents |
| Planning | Agents |
| Memory | Agents |
| Scratchpad | Agents |
| Multi-agent system | Agents |
| Orchestrator | Agents |
| Worker agent | Agents |
| Handoff | Agents |
| Guardrail | Agents |
| Human in the loop | Agents |
| Autonomy | Agents |
| Task decomposition | Agents |
| Reflection | Agents |
| Self-critique | Agents |
| Self-consistency | Agents |
| Tree of thoughts | Agents |
| Graph of thoughts | Agents |
| Workflow | Agents |
| State machine | Agents |
| Sandbox | Agents |
| Browser agent | Agents |
| Computer use | Agents |
| Code interpreter | Agents |
| MCP | Agents |
| Model Context Protocol | Agents |
| Agent evaluation | Agents |
| Trajectory | Agents |
| Alignment | Safety and alignment |
| Misalignment | Safety and alignment |
| Jailbreak | Safety and alignment |
| Prompt injection | Safety and alignment |
| Indirect prompt injection | Safety and alignment |
| Guardrail | Safety and alignment |
| Content filter | Safety and alignment |
| Moderation | Safety and alignment |
| Red teaming | Safety and alignment |
| Adversarial prompt | Safety and alignment |
| Hallucination | Safety and alignment |
| Confabulation | Safety and alignment |
| Sycophancy | Safety and alignment |
| Bias | Safety and alignment |
| Fairness | Safety and alignment |
| Toxicity | Safety and alignment |
| Privacy | Safety and alignment |
| PII leakage | Safety and alignment |
| Membership inference | Safety and alignment |
| Data poisoning | Safety and alignment |
| Backdoor | Safety and alignment |
| Model extraction | Safety and alignment |
| Prompt leakage | Safety and alignment |
| Sandbox escape | Safety and alignment |
| Capability evaluation | Safety and alignment |
| Dangerous capability | Safety and alignment |
| Dual use | Safety and alignment |
| Model card | Safety and alignment |
| System card | Safety and alignment |
| Responsible scaling policy | Safety and alignment |
| GPU | Hardware and infrastructure |
| TPU | Hardware and infrastructure |
| NPU | Hardware and infrastructure |
| LPU | Hardware and infrastructure |
| HBM | Hardware and infrastructure |
| VRAM | Hardware and infrastructure |
| Memory bandwidth | Hardware and infrastructure |
| Tensor core | Hardware and infrastructure |
| CUDA | Hardware and infrastructure |
| ROCm | Hardware and infrastructure |
| Kernel | Hardware and infrastructure |
| Compute capability | Hardware and infrastructure |
| FLOPS | Hardware and infrastructure |
| TOPS | Hardware and infrastructure |
| Interconnect | Hardware and infrastructure |
| NVLink | Hardware and infrastructure |
| InfiniBand | Hardware and infrastructure |
| Cluster | Hardware and infrastructure |
| Node | Hardware and infrastructure |
| Rack | Hardware and infrastructure |
| Wafer-scale engine | Hardware and infrastructure |
| SRAM | Hardware and infrastructure |
| Cache hierarchy | Hardware and infrastructure |
| PCIe | Hardware and infrastructure |
| Power envelope | Hardware and infrastructure |
| Thermal design power | Hardware and infrastructure |
| Utilization | Hardware and infrastructure |
| Model serving | Hardware and infrastructure |
| Inference server | Hardware and infrastructure |
| Autoscaling | Hardware and infrastructure |
| Cold start | Hardware and infrastructure |
| Edge inference | Hardware and infrastructure |
| API | Business and pricing |
| Endpoint | Business and pricing |
| Rate limit | Business and pricing |
| Quota | Business and pricing |
| Token pricing | Business and pricing |
| Input tokens | Business and pricing |
| Output tokens | Business and pricing |
| Cached tokens | Business and pricing |
| Blended price | Business and pricing |
| Price per million tokens | Business and pricing |
| Context pricing | Business and pricing |
| Batch API | Business and pricing |
| Provisioned throughput | Business and pricing |
| Pay as you go | Business and pricing |
| Subscription | Business and pricing |
| Seat pricing | Business and pricing |
| Enterprise contract | Business and pricing |
| SLA | Business and pricing |
| Uptime | Business and pricing |
| Latency budget | Business and pricing |
| Cost per task | Business and pricing |
| Cost per query | Business and pricing |
| Gross margin | Business and pricing |
| Inference cost | Business and pricing |
| Training cost | Business and pricing |
| Total cost of ownership | Business and pricing |
| Vendor lock-in | Business and pricing |
| Data residency | Business and pricing |
| Compliance | Business and pricing |
| Procurement | Business and pricing |
| Multimodal model | Multimodal |
| Vision-language model | Multimodal |
| Image encoder | Multimodal |
| Vision transformer | Multimodal |
| Patch embedding | Multimodal |
| Image token | Multimodal |
| OCR | Multimodal |
| Document understanding | Multimodal |
| Chart understanding | Multimodal |
| Video understanding | Multimodal |
| Audio model | Multimodal |
| Speech-to-text | Multimodal |
| Text-to-speech | Multimodal |
| Voice cloning | Multimodal |
| Image generation | Multimodal |
| Diffusion model | Multimodal |
| Latent diffusion | Multimodal |
| Text-to-image | Multimodal |
| Image editing | Multimodal |
| Video generation | Multimodal |
| Dataset | Data |
| Corpus | Data |
| Data license | Data |
| Open weights | Data |
| Proprietary model | Data |
| Data card | Data |
| Annotation | Data |
| Labeling | Data |
| Human feedback | Data |
| Preference pair | Data |
| Synthetic dataset | Data |
| Data augmentation | Data |
| Data contamination | Data |
| Benchmark dataset | Data |
| Training corpus | Data |
| Web crawl | Data |
| Common Crawl | Data |
| Data governance | Data |
| Data retention | Data |
| Data provenance | Data |
| Transformer explained | Architecture |
| Attention explained | Architecture |
| Self-attention explained | Architecture |
| Multi-head attention explained | Architecture |
| Feed-forward network explained | Architecture |
| Layer normalization explained | Architecture |
| Residual connection explained | Architecture |
| Positional encoding explained | Architecture |
| Rotary positional embedding explained | Architecture |
| Mixture of experts explained | Architecture |
| Sparse mixture of experts explained | Architecture |
| Dense model explained | Architecture |
| Decoder-only model explained | Architecture |
| Encoder-decoder model explained | Architecture |
| Embedding layer explained | Architecture |
| Output head explained | Architecture |
| Vocabulary explained | Architecture |
| Hidden state explained | Architecture |
| Activation function explained | Architecture |
| GELU explained | Architecture |
| SwiGLU explained | Architecture |
| Softmax explained | Architecture |
| Logits explained | Architecture |
| Autoregressive model explained | Architecture |
| Context window explained | Architecture |
| Attention mask explained | Architecture |
| Causal mask explained | Architecture |
| Grouped-query attention explained | Architecture |
| Multi-query attention explained | Architecture |
| Flash attention explained | Architecture |
| Sliding window attention explained | Architecture |
| State space model explained | Architecture |
| Recurrent layer explained | Architecture |
| Convolutional layer explained | Architecture |
| Normalization layer explained | Architecture |
| Dropout explained | Architecture |
| Weight tying explained | Architecture |
| Parameter count explained | Architecture |
| Model size explained | Architecture |
| Base model explained | Architecture |
| Pre-training explained | Training |
| Fine-tuning explained | Training |
| Instruction tuning explained | Training |
| Supervised fine-tuning explained | Training |
| RLHF explained | Training |
| DPO explained | Training |
| PPO explained | Training |
| GRPO explained | Training |
| Reward model explained | Training |
| Preference data explained | Training |
| Synthetic data explained | Training |
| Data curation explained | Training |
| Data cleaning explained | Training |
| Deduplication explained | Training |
| Tokenization explained | Training |
| Byte-pair encoding explained | Training |
| SentencePiece explained | Training |
| Learning rate explained | Training |
| Batch size explained | Training |
| Gradient accumulation explained | Training |
| Gradient checkpointing explained | Training |
| Mixed precision explained | Training |
| LoRA explained | Training |
| QLoRA explained | Training |
| Adapter explained | Training |
| Prefix tuning explained | Training |
| Prompt tuning explained | Training |
| Distillation explained | Training |
| Quantization explained | Training |
| Pruning explained | Training |
| Knowledge distillation explained | Training |
| Continual learning explained | Training |
| Catastrophic forgetting explained | Training |
| Curriculum learning explained | Training |
| Reinforcement learning explained | Training |
| Chain-of-thought training explained | Training |
| Self-play explained | Training |
| Constitutional AI explained | Training |
| Red teaming explained | Training |
| Data mixture explained | Training |
| Inference explained | Inference |
| Token explained | Inference |
| Prompt explained | Inference |
| Completion explained | Inference |
| Temperature explained | Inference |
| Top-p explained | Inference |
| Top-k explained | Inference |
| Beam search explained | Inference |
| Greedy decoding explained | Inference |
| Sampling explained | Inference |
| KV cache explained | Inference |
| Paged attention explained | Inference |
| Speculative decoding explained | Inference |
| Batching explained | Inference |
| Continuous batching explained | Inference |
| Throughput explained | Inference |
| Latency explained | Inference |
| Time to first token explained | Inference |
| Tokens per second explained | Inference |
| Streaming explained | Inference |
| Stop sequence explained | Inference |
| Logit bias explained | Inference |
| Repetition penalty explained | Inference |
| Frequency penalty explained | Inference |
| Presence penalty explained | Inference |
| Context length explained | Inference |
| Max output tokens explained | Inference |
| Truncation explained | Inference |
| Chunking explained | Inference |
| Prompt caching explained | Inference |
| Prefix caching explained | Inference |
| Quantized inference explained | Inference |
| FP8 inference explained | Inference |
| INT4 inference explained | Inference |
| GPU memory explained | Inference |
| VRAM explained | Inference |
| Model parallelism explained | Inference |
| Tensor parallelism explained | Inference |
| Pipeline parallelism explained | Inference |
| Expert parallelism explained | Inference |
| Benchmark explained | Evaluation |
| Leaderboard explained | Evaluation |
| Evaluation harness explained | Evaluation |
| Held-out set explained | Evaluation |
| Test set explained | Evaluation |
| Validation set explained | Evaluation |
| Contamination explained | Evaluation |
| Data leakage explained | Evaluation |
| Few-shot evaluation explained | Evaluation |
| Zero-shot evaluation explained | Evaluation |
| Chain-of-thought evaluation explained | Evaluation |
| Pass at k explained | Evaluation |
| Exact match explained | Evaluation |
| F1 score explained | Evaluation |
| BLEU explained | Evaluation |
| ROUGE explained | Evaluation |
| Perplexity explained | Evaluation |
| Calibration explained | Evaluation |
| Arena Elo explained | Evaluation |
| Pairwise preference explained | Evaluation |
| Human evaluation explained | Evaluation |
| LLM-as-a-judge explained | Evaluation |
| Rubric explained | Evaluation |
| Regression test explained | Evaluation |
| Eval suite explained | Evaluation |
| Task suite explained | Evaluation |
| Long-context evaluation explained | Evaluation |
| Multimodal evaluation explained | Evaluation |