context caching
optimization
Context caching is the serving optimization that reuses previously processed prompt context state to avoid recomputing identical prefixes - it is a major latency and cost lever…
2,848 technical terms and definitions
optimization
Context caching is the serving optimization that reuses previously processed prompt context state to avoid recomputing identical prefixes - it is a major latency and cost lever…
dialogue
Context carryover is the ability of a dialogue system to maintain and utilize information from previous conversation turns when processing new user messages. It is fundamental…
llm optimization
Context Compression is the technique for reducing the effective length of input sequences while preserving semantic information essential for language model reasoning — Context…
prompting
Context compression techniques is the set of methods that reduce token footprint of prompts while preserving critical semantic content - compression enables larger effective…
prompting techniques
Context Distillation is a method that transfers behavior from long-context prompting into a compact model or shorter prompt form - It is a core method in modern LLM execution…
Context Encoder Inpainting Introduction & Motivation Context Encoder Inpainting addresses a central problem in selfsupervised and inpainting-region-selection architecture: how…
architecture
Context extension techniques is the methods that increase effective model context usage through positional scaling, sparse attention, memory compression, or retrieval planning…
long context llm, rope scaling, long sequence, 128k context
Context Length Extension is the set of techniques for enabling LLMs trained on short sequences to process much longer sequences at inference time — expanding usable context…
challenges
The context window is the maximum amount of text — measured in tokens, not words — that a language model can attend to at once. It is the model's working memory: the prompt you…
rag
Context ordering is the strategy for sequencing retrieved chunks within the prompt to maximize evidence utility and minimize positional degradation - ordering determines which…
rag
Context Ordering is the strategy of arranging retrieved chunks to maximize model attention and answer reliability - It is a core method in modern RAG and retrieval execution…
llm architecture
Context overflow occurs when input exceeds a language model's maximum context window (token limit), requiring truncation, chunking, or summarization strategies to fit within…
distributed training
Context Parallelism is a distributed training and inference strategy that partitions long input sequences across multiple GPUs — enabling processing of context lengths (100K-1M+…
rag
Context placement is the decision of where retrieved evidence is inserted within the prompt relative to instructions, conversation history, and user query - placement affects how…
rag
Context Precision is the proportion of retrieved context that is actually relevant to the target answer - It is a core method in modern RAG and retrieval execution workflows. What…
self-supervised learning
Context Prediction is a self-supervised pretext task where the model predicts the spatial relationship between two image patches — given a center patch and a neighboring patch…
self-supervised learning
Context prediction pretext learning is the task of predicting relative spatial position between image patches to force models to learn object layout and scene geometry…
prompting
Context pruning is the selective removal of low-value prompt content to maximize useful information density within limited context windows - it helps maintain performance as…
rag
Context Pruning is the removal of low-value tokens or passages from context windows before generation - It is a core method in modern RAG and retrieval execution workflows. What…
rag
Context Recall is the extent to which retrieved context contains the information required to produce the correct answer - It is a core method in modern RAG and retrieval execution…
rag
Context relevance is the degree to which retrieved passages are directly useful for answering the current user query - it measures retrieval quality before generation quality can…
rag
Context Relevance is the degree to which retrieved context is useful for answering the specific query - It is a core method in modern RAG and retrieval execution workflows. What…
context switch overhead, context switch, switch overhead
Context Switch Overhead Introduction & Motivation Context Switch Overhead addresses a central problem in systems process control and context-switch-latency-penalty prediction: how…
context length, context size, context limit, maximum context, max context tokens, window size, max tokens, context window size
The context window is the maximum amount of text — measured in tokens, not words — that a language model can attend to at once. It is the model's working memory: the prompt you…
llm architecture
The context window is the maximum amount of text — measured in tokens, not words — that a language model can attend to at once. It is the model's working memory: the prompt you…
truncate, summarize
Context Window Management is the set of strategies for efficiently utilizing a language model's fixed token limit across system prompts, conversation history, retrieved documents…
prompting
Context window management is the process of controlling what information is included in each model call to stay within token limits while preserving task-critical context - it…
optimization
Context Window Management is the strategy for fitting relevant information within a model's maximum context length - It is a core method in modern semiconductor AI serving…
audio & speech
ContextNet is a convolution-based speech-recognition architecture designed to capture long context with efficient temporal processing - Stacked context modules aggregate broader…
Contextual Anomaly Detection Introduction & Motivation Contextual Anomaly Detection addresses a central problem in anomaly detection and {target} scoring: how to Detect {target}…
advanced training
Contextual augmentation is a data-augmentation approach that creates training samples using context-preserving transformations - Augmentation operators rewrite or perturb examples…
reinforcement learning
A contextual bandit is a reinforcement learning framework where an agent makes decisions based on context (features/state) available at decision time, receives a reward for its…
recommendation systems
A contextual bandit is a reinforcement learning framework where an agent makes decisions based on context (features/state) available at decision time, receives a reward for its…
Contextual Multi-Armed Bandits with LinUCB Introduction & Motivation Contextual Multi-Armed Bandits with LinUCB addresses a central problem in recsys systems and {target}…
Contextual Bandit Learning Introduction & Motivation Contextual Bandit Learning addresses a central problem in reinforcement policy learning and context-utilization-quality…
rag
Contextual Compression is a method that condenses retrieved context to only information relevant for answering the current query - It is a core method in modern RAG and retrieval…
interpretability
Contextual Decomposition is an attribution method that separates contributions of selected inputs from surrounding context - It helps explain sequence predictions by partitioning…
contextual embedding bert, contextual embedding, embedding bert
Contextual Embedding BERT Introduction & Motivation Contextual Embedding BERT addresses a central problem in nlp process control and context-representation-bidirectional…
rag
Contextual embeddings incorporate surrounding context to generate more accurate document representations. Problem: Standard chunking embeds each chunk in isolation, losing…
Contextual Outlier Detection with Behavioral Partitioning Introduction & Motivation Contextual Outlier Detection with Behavioral Partitioning addresses a central problem…
Contextual Recommendation Systems Introduction & Motivation Contextual Recommendation Systems addresses a central problem in recsys systems and {target} optimization: how to Build…
contextual word embedding elmo, contextual word embedding, word embedding elmo, contextual word, word embedding, embedding elmo
Contextual Word Embedding ELMo Introduction & Motivation Contextual Word Embedding ELMo addresses a central problem in nlp process control…
quality & reliability
Contingency Table is a cross-tabulated count matrix summarizing joint frequency of categorical variables - It is a core method in modern semiconductor statistical experimentation…
learning, catastrophic, forgetting, lifelong, learning, replay, consolidation
Continual Learning Catastrophic Forgetting is training neural networks sequentially on tasks without forgetting previously learned tasks, addressing the catastrophic forgetting…
Continual Active Feature Acquisition under Concept Drift Introduction & Motivation Continual Active Feature Acquisition under Concept Drift addresses a central problem…
Lifelong Bleed Air Valve Degradation in Commercial Fleets Introduction & Motivation Lifelong Bleed Air Valve Degradation in Commercial Fleets addresses a central problem…
Incremental One-Class SVM with Boundary Expansion Introduction & Motivation Incremental One-Class SVM with Boundary Expansion addresses a central problem in continual…
Lifelong Dissolved Oxygen Forecasting in Fish Farms Introduction & Motivation Lifelong Dissolved Oxygen Forecasting in Fish Farms addresses a central problem in continual…
Incremental Audio Event Classification with Sound Replay Introduction & Motivation Incremental Audio Event Classification with Sound Replay addresses a central problem…
Continual Visual Servo Docking in Variable Harbors Introduction & Motivation Continual Visual Servo Docking in Variable Harbors addresses a central problem in continual…