Индекс терминов
Индекс терминов
Термины автоматически собраны из вопросов, ответов, практик и ТЗ.
Всего терминов: 7743
A
- A* search
- A/B test
- A/B testing
- A/B тест метрик качества
- A/B тестирование промптов
- A100
- A100 80GB
- A10G
- A2A
- A2A Protocol
- A2C
- A3C
- Aaronson
- AARRR metrics
- ABC
- ablation study
- Abort
- Absolute Positional Encoding
- ABSTAIN
- Abstraction layer
- abstractive summarization
- Accelerate
- acceptance rate
- acceptance threshold
- acceptance window
- Access control
- accessibility tree
- accidental harmful actions
- accumulation steps
- accuracy
- accuracy drop
- Accuracy on goldenset
- accuracy предсказания winner
- ACID
- ACID транзакции
- ACK
- acks=all
- ACL
- ACME
- acquisition function
- action
- Action Correctness
- Action Distribution Drift
- Action F1
- action head
- action items
- Action safety rate
- actions/github-script
- activation offloading
- Activation patching
- Activation quantization
- Activation Statistics
- Activation steering
- activation variance
- activations
- Active Connections
- active learning
- active learning loop
- Active Probing
- Active-Active архитектура
- Active-passive
- Active-Passive архитектура
- Activity
- Actix
- Actor Model
- Actor-Critic architecture
- ActorSystem
- AdaBelief
- AdaGrad
- AdaLoRA
- Adam optimizer
- AdamW
- adapter conflict
- adapter conflicts
- Adapter layers
- adapter_config.json
- adapter_model.safetensors
- AdapterFusion
- adapters
- Adaptive backoff
- Adaptive buffering
- Adaptive computation time
- adaptive compute
- Adaptive concurrency
- Adaptive context
- Adaptive decomposition
- Adaptive design
- adaptive KL controller
- Adaptive KL penalty
- adaptive learning rates
- Adaptive Prompting
- Adaptive RAG
- adaptive rate limiting
- adaptive reasoning depth
- adaptive resource allocation
- Adaptive Retrieval
- adaptive routing
- Adaptive sampling
- adaptive sparse attention
- Adaptive Wave Decoding
- Adaptive-RAG
- Add
- add_messages
- additionalProperties
- Additive attention
- Additive masking
- Additive Quantization
- ADF test
- Adjudication
- AdmissionController
- Advantage
- advantage estimation
- AdvBench
- Adversarial attacks
- adversarial examples
- Adversarial Examples for Code
- adversarial filtering
- Adversarial generation
- adversarial hard negative
- adversarial input
- Adversarial Instructions
- adversarial patch
- Adversarial pattern
- Adversarial POPE
- adversarial probing
- adversarial prompt detection
- Adversarial prompts
- Adversarial query
- Adversarial reprogramming
- adversarial retrieval
- Adversarial suffix
- adversarial training
- AdvGLUE
- AES-256
- Affine Transformation
- AG News
- Age самого старого сообщения
- agent
- Agent Card
- Agent Communication Protocol
- agent distillation
- agent explanation fidelity
- Agent Framework
- agent handover
- Agent looping
- Agent permissions
- Agent Pipeline
- agent registry
- Agent safety constraints
- Agent self-confidence
- agent specification
- Agent state
- agent state management
- agent swarm
- agent system
- Agent tools
- Agent utilization
- agent versioning
- Agent with Memory
- Agent with tools
- Agent-based approach
- Agent-Eval
- agent-manager
- Agent-Validator
- AgentBench
- AgentContext
- AgentCostTracker
- AgentError
- AgentExecutor
- Agentic AI
- Agentic chunking
- Agentic loops
- agentic observability
- Agentic planning
- Agentic RAG
- agentic workflows
- AgentInterface
- AgentManifest
- AgentOps
- AgentPool
- AgentRunner
- AgentScope
- aggregation
- Aging
- Agno
- Agreement
- agreement matrix
- AI agents
- AI Feynman
- AI Verify
- AI-constructed formal languages
- AIC
- Aider
- AIM
- aio_pika
- aiobreaker
- aiocache
- aiohttp
- aiokafka
- aiomonitor
- aioresponses
- aiortc
- aiosmtpd
- Airbyte
- Airflow
- ajv
- Akka
- ALBEF
- ALBERT
- ALCE
- Aleatoric uncertainty
- Alembic
- alembic upgrade head
- alert
- Alert rule
- alert rules
- Alert threshold
- alerting
- alerting rule
- Alertmanager
- ALFWorld
- algbw
- ALiBi
- ALIGN
- Alignment budget
- alignment tax
- ALiPy
- all-mpnet-base-v2
- all-to-all communication
- all_reduce_perf
- AllGather
- allkeys-lfu
- allkeys-lru
- Allocate
- allocated_bytes / reserved_bytes
- Allocation rule
- allowed variations
- allowed_patterns
- AllReduce
- allreduce_perf
- Allure
- Alpaca
- Alpaca-format
- Alpaca-LoRA
- AlpacaEval
- alpha
- AlphaFold 3
- AlphaGo
- AlphaProof
- AlphaSearch
- AlphaZero
- alpine
- altinity-clickhouse-grafana plugin
- Amazon Comprehend
- Amazon EMR
- Amazon Kinesis
- Amazon Mechanical Turk
- Amazon Reviews
- Amazon SageMaker Ground Truth
- Amazon Step Functions
- ambiguity_score
- ambiguous
- ambiguous queries
- Ambiguous query
- AMD EPYC
- AMD MI300X
- Amino acid sequence
- amortized upfront
- Amplitude
- AMQP
- Analytics events
- anchoring
- anchoring bias
- anisotropic quantization
- ANLI
- ANN
- ANN индекс
- ANN-benchmarks
- annotator
- annotator calibration
- Annoy
- Anomaly Detection
- Anonymity
- Anonymization by prompting LLM
- Anonymized data
- ANSI-коды
- Answer
- Answer quality
- Answer Recall
- Answer relevance
- answer_correctness
- answer_exact_match
- AnswerVerifier
- Ant Colony Optimization
- Anthropic Claude API
- Anthropic evals
- Anthropic HH-RLHF
- Anthropic prompt caching
- Anthropic SDK
- anti-contamination
- any-to-any generation
- Anycast
- anytree
- AOF
- AOF rewrite
- AOT compilation
- AP
- Apache 2.0
- Apache Atlas
- Apache Beam
- Apache Bench
- Apache Flink
- Apache Spark Streaming
- Apache TVM
- API
- API access control
- API call
- API contract
- API costs
- API error
- API key
- API key rate limiting
- API tokens
- API вызовы инструментов
- Apicurio
- Apicurio Registry
- APM
- AppArmor
- Append-only log
- appendfsync
- Approval latency
- Approval rate
- Approval voting
- approve/deny
- approximate LFU
- APScheduler
- AQLM
- AQuA
- arbitrary resolution
- ARC
- ARC-AGI
- ARC-Challenge
- Architecture rules
- Argilla
- Argo Rollouts
- ArgoCD
- argparse
- Argument Accuracy
- ARIMA
- Arithmetic intensity
- Arize
- Arize AI
- Arize Phoenix
- ARM Neoverse V2
- ARPC
- ARPU
- Array of Strings
- Arrival time
- artifact
- Artificial Analysis
- ASCII art
- ASE
- ASQA
- Assertions
- assertions_feedback
- assertions_handler
- assertions_max_retries
- Assistants API
- associative memory
- Assumptions
- AST
- Asymmetric Distance Computation
- ASYNC
- async call
- async copy
- async CUDA
- async data movement
- async GEMM
- async generator
- async messaging
- async response
- async session
- async with
- async-profiler
- async/await
- AsyncAPI
- asynchronous data copy
- Asynchronous Execution
- Asynchronous H2D copy
- asynchronous preprocessing
- Asynchronous SM-to-SM copy
- asynchronous transaction barriers
- Asynchronous verification
- asyncio
- Asyncio timeouts
- asyncio.Barrier
- asyncio.gather
- asyncio.Lock
- asyncio.Queue
- asyncio.Semaphore
- asyncio.sleep
- asyncio.Task
- asyncio.wait_for
- asyncpg
- AsyncPipeline
- at-least-once semantics
- At-most-once
- at-most-once семантика
- Atheris
- atomic action
- atomic append
- Atomic Operation
- atomic operations
- atomic read-write
- attack success rate
- Attention
- attention compression
- Attention dilution
- Attention dropout
- attention entropy
- attention fusion
- Attention heads
- Attention kernel
- attention masking
- attention metrics
- attention normalization
- attention pattern analysis
- Attention patterns
- attention projections
- Attention pruning
- Attention score
- attention sink
- Attention: PV
- Attention: QK^T
- Attribution-perturbation consistency score
- AUC
- Auction
- auction-based task allocation
- Auctioneer
- audience calibration
- Audio encoder
- Audio RAG
- AudioCraft
- AudioLM
- Audit logging
- Authentication
- Author
- Authority score
- Auto Scaling
- auto-commit
- Auto-docs
- auto-gptq
- auto-instrumentation
- auto-merging retrieval
- Auto-remediation
- auto-scaling
- Auto-success rate
- auto-tuning
- auto_wrap_policy
- AutoAWQ
- Autocut
- AutoDAN
- autoencoder
- AutoGen
- autogenerate
- AutoGPT
- AutoGPTQForCausalLM
- Autograd
- automated testing
- automatic baseline computation
- automatic labeling
- Automatic Prompt Engineering
- AutoModelForCausalLM
- Autoregressive
- autoregressive generation
- Autoregressive inference
- autoregressive model
- autoscaling inference
- AutoScheduler
- AutoTokenizer
- AutoTVM
- auxiliary loss
- availability
- available blocks
- Average cost per delegation
- average handling time
- Average handoffs per query
- Average iterations
- Average Pairwise Similarity
- Average steps
- Average steps per milestone
- Average Wait Time
- AverageUtilization
- averageValue
- Avro
- AvroConsumer
- AvroProducer
- AWQ
- AWS capacity reservations
- AWS CLI
- AWS CloudWatch
- AWS Cost and Usage Report
- AWS DMS
- AWS EC2
- AWS Global Accelerator
- AWS Glue
- AWS Glue Schema Registry
- AWS KMS
- AWS Price List API
- AWS Pricing Calculator
- AWS Region
- AWS Secrets Manager
- AWS SQS
- Axolotl
- Azure AI Red Team Tools
- Azure Content Safety
- Azure Durable Functions
- Azure Key Vault
- Azure Monitor
B
- B-tree
- B200
- B2B
- B2C
- BAAI/bge-large-en
- BAAI/bge-m3
- bAbI
- BabyAGI
- back-translation
- backdoor
- Backdoor poisoning
- backdoor watermarking
- Backend Engineer
- Backfill
- Background task
- BackgroundTasks
- Backlog
- backpressure
- backpropagation
- Backtranslation
- backup
- backward compatibility
- Backward generation
- backward pass
- BAE
- Bag-of-Words
- Bag-of-words bias
- Baggage
- Balance coefficient beta
- Balance factor
- bandit
- bank conflicts
- bare-metal инстанс
- Base frequency
- base metrics
- Base64 encoding
- BaseAgent
- BaseCallbackHandler
- baseline
- Baseline Scenario
- Baseline utilisation
- Batch Encoding
- Batch Hard Triplet Mining
- Batch inference
- Batch ingestion
- batch matrix multiplication
- batch mix
- Batch mode
- Batch RAG
- batch search
- batch size
- batch update
- batch write
- Batch-запрос
- batch-операции
- batch/v1 Job
- batched scoring
- Batching scheduler
- Batching timeout
- Batching tool calls
- BatchNorm
- BatchSpanProcessor
- Bayesian approximation
- Bayesian Elo
- Bayesian optimization
- BBH
- BBQ
- beam search
- beam_width
- BeautifulSoup
- BeeGFS
- behavior cloning
- Behavior Drift
- Behavioral profiling
- Behavioral testing
- BEIR
- Belief Tracking
- benchmark
- benchmark chasing
- benchmark overfitting
- benchmark task generation
- Benchmarks
- Benign prompt
- Benjamini-Hochberg
- Bernoulli distribution
- BERT
- BERT classifier
- BERT-Attack
- BERT-large
- BERT-masking
- BERT-tiny
- BERTopic
- BERTscore
- BertViz
- beta
- BF16
- bge-large-en-v1.5
- BGE-reranker
- bgsave
- Bi-encoder
- bias
- bias amplification
- Bias Rate
- BIC
- Bid
- Bidirectional LSTM
- BIG-bench
- BigBird
- BigQuery
- Binary classifier
- Binary Cross-Entropy Loss
- binary good/bad
- binary metric
- Binary quantization
- binary search
- bind mount
- Binding
- binning
- binomial testing
- Binpacking
- BioMedLM
- biometric features
- biometric identification
- bit masks
- Bit signature
- bitarray
- bitsandbytes
- bitsandbytes 4-bit quantization
- black
- black-box
- Black-box attack
- Black-box extraction
- Black-box watermarking
- Blackboard
- Blackbox Exporter
- blacklist
- Blacklist/Whitelist
- Blackwell architecture
- Blame attribution problem
- Blameless culture
- Blameless postmortem
- BLAS
- BLEU
- BLEU-4
- BLEURT
- BLIP
- BLIP-2
- Blob storage
- block
- block allocation
- Block manager
- Block-based allocation
- Block-sparse attention
- block_size
- blocking cases
- Blocksworld
- Blockwise Parallel Transformer
- Bloom filter
- Bloom filter parameters
- BLPOP
- blue team
- Blue-green deployment
- BM25
- BM25 hard negative
- BNS
- BOM
- bonding
- Bonferroni correction
- Boolean Filters
- BoolQ
- bootstrap
- Bootstrap estimation
- BootstrapFewShot
- BootstrapFewShotWithRandomSearch
- Borda count
- BOS token
- boto3
- Bottleneck
- bottlenecks
- boundaries
- bounded queue
- Bounded rationality
- BoundedSemaphore
- bounding box coordinates
- bounding boxes
- boxplot
- BPE
- BPTT
- Bradley-Terry model
- Bradley-Terry модель
- branch coverage
- branch efficiency
- branch prediction
- Branch protection
- branch references
- branch rules
- Branching
- breadth-first traversal
- Break-even Chart
- break-even point
- Breaking changes
- Brent method
- bridge
- BridgeTower
- Brier score
- Broadcast
- browser agent
- brute force
- bucket
- Bucket resolution
- bucketing
- budget balance
- budget monitoring
- budget per session
- budget usage
- Budget utilization
- BudgetExceededError
- budgeting
- buffer management
- BufferWindowMemory
- Build engine
- build time
- building index
- Bulk API
- bulk insert
- bulkhead
- Bully algorithm
- Bulyan
- bunched kernel launches
- burn rate
- burst
- burst allowance
- Bus utilization
- busbw
- BYOC
- ByT5
- byte-level tokenization
- bytearray
C
- CAC
- Cache Agent
- Cache effect
- cache entry
- cache eviction policies
- Cache hit ratio
- cache invalidation
- cache invalidation strategies
- cache miss
- Cache misses
- cache prefix
- cache rollback
- Cache stability
- Cache stampede
- Cache Systems
- cache warming
- Cache-Aside
- Cache-Control
- cache_control
- cache_creation_input_tokens
- cache_key
- cache_read_input_tokens
- cached response
- CachedContent
- CacheInterface
- cachetools
- caching
- Caching decorator
- Caching popular vectors
- Caching strategies in AI systems
- CAF
- CalibratedClassifierCV
- Calibration
- calibration dataset
- calibration error
- Calibration queries
- Calibration RM
- call directive
- Call graph
- call-center аналитика
- Call-level scoping
- callback
- CallbackManager
- callbacks
- calls per session
- Camelot
- Camunda
- canary deployment
- canary examples
- canary:true
- Cancellation
- Cancellation token
- cancellation_latency
- cancellation_rate
- CancelledError
- candidate
- candidate tree
- canonical perturbations
- CAP theorem
- CAP_NET_RAW
- Capability
- capability negotiation
- Capability-based negotiation
- Capacity
- capacity factor
- capacity planning
- CapEx
- caption generation
- Caption-based approach
- Captum
- cardinality
- carryover effect
- CAS
- cascade
- cascade failure
- Cascade Uncertainty Amplification
- Cascading
- cascading agent system
- cascading agent systems
- cascading failures
- Cassandra
- catastrophic forgetting
- Causal attention
- Causal density
- causal LM
- Causal LM Head
- causal masking
- causal reasoning
- Causal Tracing
- causal-conv1d
- causalnex
- CDC
- CDF
- CDN
- CDNA3
- Celery
- Celery beat
- Centering
- central planner
- Central tendency
- central tendency bias
- Centralized architecture
- centroid
- Certified robustness
- CFD
- cgroups
- chain
- chain decomposition
- chain of actions
- Chain of Responsibility
- Chain rule
- Chain-of-Thought
- Chain-of-Thought fine-tuning
- Chain-of-Thought generation
- Chain-of-Thought критика
- Chain-of-verification
- CHAIR
- CHAIRi
- CHAIRs
- Chameleon
- Chameleon attack
- change detection
- channel
- chaos engineering
- Chaos Mesh
- Chaos Monkey
- Chaos Toolkit
- chaosmonkey
- ChaosProxy
- chaostoolkit
- Character Error Rate
- Character-level attack
- ChartQA
- Chat Completion
- ChatCompletion
- ChatGPT API
- ChatML
- ChatOllama
- ChatOpenAI
- ChatPromptTemplate
- Check
- check_secrets.py
- Checker
- CheckList
- Checkpoints
- chi-square test
- Chi-squared test
- Child span
- Chimera
- Chinook
- Choreography
- Choreography SAGA
- chosen/rejected pairs
- Chroma
- ChromaDB
- Chroot
- chunk enrichment
- Chunk overlap
- Chunk Recall@k
- chunk size
- chunk-based search
- chunked prefill
- Chunked synthesis
- chunking
- Chunkization
- chunks
- churn
- Churn Rate
- CI
- CI validation
- CI-артефакты
- CI-линтер
- CI/CD
- CI/CD for AI
- CI/CD for ML pipelines
- CI/CD for prompts
- CI/CD для промптов
- CIDEr
- CIF
- CIFAR-10
- Cilium
- circuit breaker
- circuitbreaker
- Citation
- citation accuracy
- citation check
- Citation checking
- CityHash
- CJK
- CKY
- claim extraction
- Claims
- CLAP
- CLARE
- Class balance
- class imbalance
- class-balanced sampling
- class_weight
- classification
- Classifier
- Classifier-Free Guidance
- Claude 3
- Claude 3 Haiku
- Claude 3 Opus
- Claude 3.5
- Claude 3.5 Sonnet
- Claude API
- Cleanlab
- CleverHans
- CLI
- click models
- ClickHouse
- clickhouse-client
- clickhouse-driver
- client output buffer
- client-side rate limiter
- Client-side rate limiting
- CLIP
- CLIP score
- Clip ε
- CLIP-based NSFW detector
- clipping
- CliRunner
- clock_gettime
- Clone
- Clone-Structured Causal Graphs
- closed-form expression
- closed-form solution
- CloudFlare
- CloudWatch
- CLS Token
- Clumsy
- cluster
- Cluster autoscaler
- Cluster ratio
- CLUSTER SETSLOT
- cluster state
- cluster-based randomization
- Clustering
- ClusterIP
- ClusterQueue
- ClusterRole
- ClusterSecretStore
- CLUTRR
- CNN
- co-adaptation
- co-shag
- coalesced_group
- CockroachDB
- COCO
- COCO API
- COCO Captions
- COCONUT
- Code
- Code Agents
- Code as Representation
- Code as Representation Language
- Code by Zapier
- Code Classification
- Code Clone Detection
- code correctness rate
- code coverage
- code coverage metrics
- code embeddings
- Code execution
- code generation
- code injection
- code review
- Code Summarization
- Code-as-Thought
- CodeBERT
- CodeBLEU
- codebook
- CodeGraph
- CoDel
- CodeSearchNet
- Codex
- Coefficient of redundancy
- Cognitive bias
- Cognitive scaffolding
- cognitive schema
- COGS
- CogVLM
- Cohen's Kappa
- Cohere Embed
- Cohere multilingual
- Cohere rerank
- cohere/embed-multilingual-v3
- coherence
- coherence illusion
- cohesion
- Colang
- ColBERT
- ColBERT multilingual
- colbert-ai
- ColBERT-v2
- ColBERTv2
- cold cache
- Cold queries
- cold standby
- Cold storage
- Cold-start
- collaboration count
- collaboration_latency_ms
- collaboration_success_rate
- collaboration_total
- Collaborative
- collapse
- collate function
- collection
- Collection per tenant
- CollNet
- Collusion
- colorama
- Colossal-AI
- Column-wise
- combinatorial auction
- COMET
- Command R+
- Commit log
- commit offset
- Commit-reveal scheme
- commit_transaction
- commitment loss
- Common Crawl
- Common item equating
- common knowledge base
- Common subexpression elimination
- common tokenizer
- CommonsenseQA
- Communication overhead explosion
- Communication rounds
- Compact+delete
- compacted topic
- Compacted topics
- Comparison Dataset
- Compensating actions
- compensating transaction
- Competitive
- compilation success rate
- Completion Queue
- completion rate
- Completion time
- completion tokens
- Complex plane rotation
- complex queries
- complexity scoring
- Compliance
- ComponentRegistry
- Composability
- composite fusion
- Composite score
- composite SLO
- Compositionality
- compound key
- compressed memory
- Compression
- Compression ratio
- Compressive Transformer
- computation graph
- compute
- compute budget
- Compute capability
- Compute costs
- Compute engine
- compute utilization
- compute-bound
- compute-communication overlap
- Compute/Communication overlap
- Compute/communication ratio
- compute_bid
- computer use agent
- Concept direction
- Concept shift
- ConceptNet
- conciseness
- concurrency
- Concurrent delegation
- Concurrent kernels
- Concurrent requests
- concurrent users
- concurrent.futures
- conda
- Conda / venv
- conditional edge
- conditional edges
- Conditional Kappa
- Conditional VAE
- conditional vector
- conditional vectors
- Conditioning
- Condorcet method
- Confidence bins
- Confidence calibration error
- Confidence penalty
- confidence score
- Confidence threshold
- confidence-based routing
- Confident learning
- Confidential computing
- CONFIG SET
- Config Versioning
- config.yaml
- CONFIG_RDMA_RXE
- ConfigMap
- configuration
- confirmation bias
- confirmation_prompt
- ConfirmDialog
- Conflict resolution
- Confluent
- Confluent Control Center
- confluent-kafka
- confluent_kafka
- Conformal prediction
- confounding
- confounding factors
- conftest.py
- Confusion matrix
- Connection Draining
- connection handling
- connection pooling
- ConnectX-6
- Consensus
- Consensus mechanism
- consistency
- Consistency checks
- Consistency Rate
- consistency regularization
- consistent hashing ring
- consistent prefix
- constant folding
- Constitution
- Constitutional adherence
- Constitutional AI
- constitutional check
- Constitutional prompt
- ConstitutionalChain
- constrained decoding
- Constrained RL
- constraint propagation
- constraint satisfaction
- constraints
- construct validity
- Consul
- Consumer
- Consumer group
- Consumer Groups
- Consumer Lag
- consumer priority
- consumer-producer
- container orchestration
- containerization
- Contamination Detection Toolkit
- Contamination rate
- Content Filter
- content hash
- content validity
- content-addressed
- content-based
- Content-Encoding
- Content-oriented application
- Content-Type: text/event-stream
- content_filter
- Context
- context adherence
- Context Builder
- Context caching
- Context Coverage
- context distillation
- context drift
- Context Engineering
- Context Extension
- Context leakage
- Context loss
- Context manager
- Context manipulation
- Context overflow
- context package
- context parallelism
- Context precision
- context preparation
- context preservation
- Context propagation
- Context Recall
- Context relevance
- context separation
- context serialization
- context truncation
- context utilization
- Context vector
- context window
- Context window explosion
- context_features
- context_length_exceeded
- context_only
- context_precision
- ContextBoundary
- contextual enrichment
- Contextual hints
- contextual representations
- Contextual retrieval
- Continuous Backup
- continuous batching
- continuous learning
- continuous monitoring
- continuous red teaming
- Continuous relaxation
- Contract testing
- contradiction
- Contradiction check
- Contradiction rate
- contrast effect
- Contrastive Activation Addition
- Contrastive decoding
- contrastive learning
- contrastive loss
- contrastive search
- CONTRIBUTING.md
- control
- Convergence
- convergence time
- ConversableAgent
- Conversation state
- Conversational repair
- ConversationBufferWindowMemory
- ConversationSummaryBufferMemory
- ConversationSummaryMemory
- Conversion Rate
- Convertible RI
- convolution
- Convoy effect
- cooldown
- Cooperative Groups
- Coordination
- Coordination Engineering
- coordination metrics
- Coordination score
- Coordinator
- COPRO
- Copy Engine
- copy with padding
- Copy-on-write
- copytest
- Coq
- core
- CoreML
- Coroutine
- Correction accuracy
- correction множественных сравнений
- Corrective RAG
- correlation
- Correlation analysis
- Correlation ID
- Correlation Metrics
- Corrupted document
- Corrupted PDF
- CORS
- Cosine Decay
- Cosine Noise Schedule
- Cosine Scheduler
- cost
- Cost Analysis
- cost anomaly detection
- cost attribution
- Cost Engineering
- cost estimator
- Cost Explorer
- cost management
- cost model
- cost of control
- cost of delegation
- cost of reasoning
- Cost optimisation
- Cost optimization
- cost penalty
- cost per 1M tokens
- Cost per agent run
- Cost per correct answer
- Cost per Delegation Path
- Cost per good answer
- cost per hour
- Cost per Improvement
- cost per request
- Cost per second of user wait
- cost per session
- Cost per successful answer
- Cost per successful task
- Cost per user
- cost per vector
- cost reduction
- cost savings
- Cost Structure
- cost tags
- cost threshold
- Cost tracking
- cost vs revenue
- Cost vs Revenue chart
- Cost-accuracy-latency trade-off
- Cost-adjusted accuracy
- cost-aware auto-scaling
- cost-aware caching
- Cost-aware planner
- cost-aware routing
- cost-latency trade-off
- cost-quality trade-off
- cost/latency/quality trade-off
- cost_table_version
- CostTracker
- Counter
- counter-offers
- Counterfactual
- Counterfactual fidelity
- counterfactual reasoning
- Covariate shift
- coverage
- coverage of API errors
- Coverage report
- Coverage ошибок
- coverage-driven generation
- coverage-guided testing
- cp.async.bulk
- cProfile
- CPU
- CPU bottleneck
- CPU inference
- CPU offload
- CPU RAM
- CPU sockets
- CPU-bound
- CPU-GPU synchronization
- CPU↔GPU transfers
- CQRS
- CRAFT
- crash rate
- Crash recovery
- crashes
- CrashLoopBackOff
- CRC errors
- CRC32
- CRDT
- Credentials
- Credit assignment
- Crescendo атака
- crew
- CrewAI
- criterion validity
- critic agent
- critical actions
- critical fraction
- Critical section
- critical workload
- Critique
- CRM
- CRNN
- Cron
- Cross-attention
- cross-contamination
- cross-correlation
- cross-correlation heatmap
- cross-encoder
- cross-encoder vs bi-encoder
- cross-encoder/nli-deberta-v3-large
- cross-entropy loss
- cross-layer attention
- cross-layer connections
- cross-lingual recall@k
- cross-lingual transfer
- cross-model
- cross-region replication
- cross-session consistency
- Cross-Session Consistency Drift
- Cross-Validation
- Cross-validation annotators
- cross-verification
- CrossEntropyLoss
- crossover
- CRUD
- Crypten
- crystal structure
- CSI
- CSV
- CSV datasource
- CTC
- CTF
- CTGAN
- CTranslate2
- CUAD
- cuBLAS
- Cuckoo filter
- CUDA
- CUDA 11.8
- CUDA API
- CUDA API calls latency
- CUDA API peer access
- CUDA caching allocator
- CUDA context
- CUDA cores
- CUDA event
- CUDA events
- CUDA Execution Provider
- CUDA graphs
- CUDA kernel
- CUDA Samples simpleP2P
- CUDA streams
- cuda-memcheck
- cuda_malloc_count
- cuda_memtest
- CUDA_VISIBLE_DEVICES
- cudaFree
- cudaLaunchCooperativeKernel
- cudaMalloc
- cudaMallocAsync
- cuDNN
- Cumulative Gain
- CUPED
- Curl
- curriculum adversarial training
- Curriculum Learning
- Curse of dimensionality
- curse of length
- Cursor
- Custom actions
- Custom CUDA kernel
- custom evaluators
- custom exporter
- custom generators
- Custom layers
- custom metric
- custom metrics
- Custom Metrics API
- Custom Resource Definition
- custom scheduler
- custom-metrics-apiserver
- CUSUM
- CuTe
- CUTLASS
- CVE
- CVSS
- Cybersecurity
- cycle detection
- Cycles
- Cyclic graph
- Cypher
- Cython
D
- d'Aspremont–Gérard-Varet mechanism
- D2
- D2H
- D3PM
- d_ff
- d_k
- d_model
- DaemonSet
- Dafny
- DAG orchestration
- Dagster
- Daily seasonality
- daily_spend_usd
- DailyDialog
- DALL-E
- damp_percent
- DAN
- dangerous action
- Dashboard
- Dask
- Data augmentation
- Data Augmentation for Code
- Data card
- data cleaning
- data collator
- Data Collection
- Data contract
- data drift
- data efficiency
- Data Engineer
- Data Exchange
- Data extraction
- Data Filtering
- Data Injection
- data labeling
- data lakehouse
- data lakes
- data lineage
- data locality
- data migration
- data mixing
- Data parallelism
- Data pipeline
- data programming
- Data Quality
- Data Quality Monitoring
- data reordering
- Data residency
- Data Sanitization
- Data staleness
- Data transfer bottleneck
- data transfers
- data types
- data validation
- Data versioning
- Data-centric AI
- Data-efficient fine-tuning
- database
- database schema
- Databricks Dolly 15k
- dataclass
- Datadog
- Datadog APM
- DataLoader
- Datalog
- DataPool
- dataset
- dataset diversity
- dataset format
- Datasketch
- dateparser
- Davies–Bouldin Index
- day-of-week effect
- DBOS
- DBpedia
- DBSCAN
- DBT
- dbt contract
- DCGM
- DCGM Exporter
- DCGM_FI_DEV_GPU_UTIL
- DCGM_FI_DEV_NVLINK_BANDWIDTH_TOTAL
- dd
- DDL
- DDoS
- dead code elimination
- Dead letter
- Dead Letter Exchange
- Dead Letter Queue
- dead neurons
- Deadband
- Deadline
- Deadlock
- deadlock detection
- deadlock detection time
- Deallocation
- DEAP
- DeBERTa-NER
- DeBERTa-NLI
- DeBERTa-v3
- Debezium
- Debugging
- decay rate
- Decentralized architecture
- decentralized control
- Decentralized system
- Decision cache
- Decision matrix
- decode
- Decoder
- Decoder-only architecture
- decoder-only LLM
- decoder-only model
- decoderbufs
- decorator
- decorator pattern
- decoupling
- Decoupling Score
- dedup table
- Deep Ensembles
- deep eval
- Deep health check
- deep learning models
- deepcopy
- deepdiff
- DeepEval
- DeepSeek V2
- DeepSeek-MoE
- DeepSeek-R1
- DeepSeek-V2
- DeepSpeech
- DeepSpeed
- DeepSpeed Inference
- DeepSpeed Pipe
- DeepSpeed-MoE
- DeepSpeed-Ulysses
- DeepStream SDK
- DeepWordBug
- Deequ
- default partition
- Default stream
- default values
- default_segment_number
- defaultdict
- Defense in Depth
- defensive distillation
- Deficit round robin
- defineTool
- Definition of Done
- degradation
- degradation detection
- Degradation slope
- degradation threshold
- Degraded mode
- degraded UX
- delayed scaling
- Delegated tools
- delegation
- delegation by exception
- delegation chain
- Delegation Efficiency
- Delegation Engineering
- delegation failure cascade
- delegation paths
- delegation_duration_seconds
- delegation_failure_total
- delegation_requests_total
- delegation_success_total
- DelegationManager
- DELETE /generations/{id
- deliberate decoding
- Deliberative consensus
- Delimiter-based approach
- Delta
- Delta Lake
- delta method
- Delta regularization
- delta weights
- demo.py
- demonstration
- Denoising
- Denoising score matching
- Dense connections
- Dense Embedding
- Dense model
- Dense rewards
- Density Functional Theory
- Deny by default
- Department
- Dependency between prompts
- dependency injection
- Dependency injection in LLM pipelines
- dependency management
- dependency tracking
- dependent functions strategy
- depends_on
- DePlot
- Deprecated
- Deprecated Field
- Depth scaling without parameters
- Depth-First Search
- dequantization
- Dequantization overhead
- Deque
- Deserialization
- Design by Contract
- Design for failure
- Deskew
- detach
- Detection Delay
- Detection LLM
- Detectron2
- determinism rate
- Deterministic runtime
- Deterministic seed
- Deterministic simulator
- Deterministic testing
- Detoxify
- DETR
- DevEx
- device_map
- DevOps Overhead
- DevTools Protocol
- DFA
- DGX
- DGX A100
- DGX H100
- Dialect
- Dialog system
- dialogue-based paradigm
- Diff
- diff2html-cli
- differencing
- differential privacy
- difflib
- diffusers
- diffusion backends
- diffusion LLM
- diffusion model
- Diffusion Models
- DiffusionBERT
- Dify
- Dify Prompt Management
- Digest
- digital signature
- Digital twin
- dilated sliding window
- diminishing returns
- DINOv2
- Direct I/O
- Direct mapping
- Direct Preference Optimization
- Directed Graph
- DirectML
- DirectoryLoader
- Disambiguation
- disaster recovery
- Disaster recovery drill
- Discounted Cumulative Gain
- Discounting
- discovery
- DiscoveryRequest
- DiscoveryResponse
- discriminated union
- Disk-based vector storage
- DiskANN
- Dispatcher
- Distilabel
- DistilBERT
- DistilGPT2
- Distinct-N
- Distractors
- distributed AI system
- distributed cache
- Distributed Data Parallel
- distributed file system
- Distributed Flash Attention
- distributed locking
- Distributed rate limiting
- distributed systems
- Distributed task queue
- Distributed tracing
- distributed training
- distributed transactions
- Distribution Collapse
- distribution fidelity
- Distroless
- divergence
- Divergent control flow
- diverse beam search
- diversity
- Diversity bonus
- diversity sampling
- divide and conquer
- Django
- DLQ count
- DLQ size
- dlt
- DMA
- DMA engine
- DNNL
- DNS failover
- DNS propagation
- DNS-based load balancing
- do_sample
- Docker
- Docker Compose
- Docker Compose networking
- Docker socket
- Docker Swarm
- docker-compose up
- docker-compose.yml
- Docker-образ
- Dockerfile
- DocLayNet
- Docling
- docstring
- DocTR
- document classification
- document injection
- Document length
- Document Loader
- document masking
- Document type
- Document Understanding
- Document-based chunking
- Document-to-version mapping
- document_id
- documents
- DOCX
- Domain
- domain expert
- domain shift
- Domain specialization
- Domain-adapted checkpoints
- Dominant Resource Fairness
- dominant strategy
- Donut
- DoRA
- DoS
- DOT format
- double auction
- double buffering
- Double Quantization
- downsampling
- downstream metrics
- Downstream model
- Downstream quality
- downstream tasks
- Downstream-процессы
- DP Inference
- DP-Fine-tuning
- DP-ретривер
- dp_accounting
- DPIA
- DPO gradient
- DPO loss
- DPOTrainer
- DQN
- draft length
- draft model
- Draft7Validator
- Dragonfly
- drain
- drain connections
- drift
- drift detection
- drift metrics
- drift retrieval-качества
- DriftDetector
- drill‑down
- Drop rate
- drop_caches
- DropConnect
- dropout
- Dry-run
- DSGE-модели
- DSL
- dslim/bert-base-NER
- DSPy
- DSPy Evaluate
- dspy.Cache
- dspy.Predict
- dspy.ProgramOfThought
- dspy.Retrieve
- dspy.serialize
- DSPyAssertionError
- dstat
- DTensor
- dtype
- Dual control
- Dual index
- dual write
- dual-write
- DuckDB
- DuckDuckGo
- duckduckgo_search
- duplicate detection
- duplicate questions
- duplicate ratio
- durable pull-подписка
- Durable state
- DVC
- DVC pipeline
- dynamic analysis
- dynamic benchmark
- dynamic benchmarks
- Dynamic confidence thresholds
- dynamic context
- Dynamic evals
- dynamic facets
- Dynamic index update
- Dynamic list
- dynamic loss scaling
- Dynamic padding
- Dynamic pricing
- dynamic programming
- Dynamic Quantization
- Dynamic range
- Dynamic representation
- Dynamic resource allocation
- Dynamic routing
- dynamic scaling
- Dynamic Scoping
- dynamic secrets
- dynamic shapes
- Dynamic Task Mapping
- dynamic temperature
- Dynamic thresholds
- dynamic tree construction
- DynamoDB
E
- E2E scenario
- E4M3
- E5-large
- e5-large-v2
- E5M2
- eager invalidation
- eager PyTorch
- EAGLE-1
- EAGLE-2
- EAGLE-3
- EAP
- Early exiting
- early fusion
- Early Stopping
- easy negative
- easy negatives
- EasyEdit
- EasyOCR
- eBPF
- EBS
- EBS volume
- EC2 instance type
- echo-сервер
- edge
- Edge Accuracy
- edge case
- edge cases
- Edge computing
- Edge deployment
- edge device
- edge_coverage
- edges
- Edit distance
- EdSurvey
- ef
- ef_construct
- ef_construction
- ef_search
- EFA
- effect size
- effective batch size
- Effective context length
- Effective cost per token
- effective reserved cost
- Efficacy
- Efficiency
- efficiency_gap
- EfficientNet
- Effort
- EFS
- EKS
- Elaboration
- ElastiCache
- Elasticity
- Elasticsearch
- ELBO
- Elbow method
- Election timeout
- ElevenLabs Turbo
- ELK
- Elo rating
- ELT
- EM convergence
- embedding
- embedding API
- Embedding Consumer
- Embedding dimension
- embedding distribution
- Embedding diversity
- Embedding drift
- Embedding dropout
- embedding inversion
- Embedding layer
- embedding model degradation
- Embedding Models
- Embedding normalization
- Embedding Pipeline
- embedding poisoning
- Embedding Rotation
- Embedding shift
- Embedding Signature
- Embedding space
- Embedding throughput
- Embedding воркеры
- Embedding-as-a-Service
- Embedding-based approach
- Embedding-based expansion
- embedding-модель
- Embeddings caching
- emergent behavior
- emergent specialization
- Empty document
- emptyDir volume
- emulator
- en_core_web_trf
- enable.idempotence
- EnCodec
- Encoder
- Encoder-decoder transformer
- Encoder-only transformer
- Encryption
- Encryption at rest
- Encryption in memory
- Encryption in transit
- Encryption in use
- encryption-at-rest
- end verifier
- End-to-end
- end-to-end compiler
- end-to-end learning
- End-to-end metrics
- End-to-end streaming
- End-to-end test
- End-to-end testing
- end-to-end обучение
- Energy distance
- energy prediction
- enforce_partition_keys
- Enron subset
- ensemble
- ensemble adversarial training
- ensemble generation
- Ensemble of models
- ensemble reward models
- ensemble RM
- ensemble-based decoding
- entailment
- Enterprise Contract
- Entity Extraction
- Entity Linking
- Entity Masking
- Entropy
- entropy bonus
- entrypoint
- enum
- env
- environment variable
- environment variables
- Envoy
- Envoy filter
- EOS token
- Ephemeral sequential node
- Episodic memory
- Epistemic uncertainty
- epistemic vigilance
- epoch
- Epsilon
- Epsilon-greedy
- Equivariance
- Equivariant GNN
- error accumulation
- error budget
- error code
- error handling
- error penalty
- Error rate
- error recurrence
- error status
- error_handling
- error_rate_429
- escalation
- Escalation flag
- Escalation of privileges
- Escalation rate
- Escalation system
- escape
- ESM3
- ESMFold
- ETag
- etcd
- ETL
- EU AI Act
- Eureqa
- Eval runner
- eval set
- eval пайплайн
- EvalAI
- Evaluate
- Evaluation
- Evaluation API
- evaluation leakage
- evaluation overfitting
- evaluation report
- Evaluator
- evaluator scores
- evaluator-based evaluation
- evaluator-based quality assessment
- evasion
- event loop
- Event loop blocking
- event pattern
- Event processing latency
- Event sourcing
- Event streaming
- event tracking
- event-driven invalidation
- Event-driven sampling
- event-stream
- EventCollector
- EventSource API
- EventSourceResponse
- eventual consistency
- Eventually consistent
- evicted keys
- Eviction policy
- Evidence Override Rate
- Evidently AI
- Evol-Instruct
- Evolution
- Evolutionary algorithms
- evolve-check
- EWC
- EWMA
- Exact attention
- Exact duplicate
- exact filter
- Exact hashing
- Exact kNN
- Exact match cache
- exact match caching
- Exact Set Match
- Exact-Match Cache
- exactly-once delivery
- examination probability
- exception class
- Exception handler
- Exceptions
- Excessive Agency
- Exchange
- Execution
- Execution Accuracy
- Execution errors
- execution feedback
- Execution guarantee
- Execution time
- executive summary
- executor agents
- EXIF
- exit 0
- ExLlama
- exllamav2
- expand
- expandable_segments
- expander
- Expansion
- Expected Calibration Error
- Expected trajectories
- expected trajectory
- expected_final
- expected_trajectory
- experience exchange
- Expert
- Expert agent
- Expert agreement
- Expert arbitration
- Expert Choice Routing
- Expert knowledge
- expert layers
- expert parallelism
- expert placement
- Expert Specialization
- Explanation Faithfulness
- Explanation-Decision Decoupling
- explicit feedback
- Explicit forget mechanism
- Explicit transitions
- exploding gradients
- Exploitation
- exploration
- Exploration vs exploitation
- exploration/exploitation
- exponential backoff
- Exponential decay initialization
- Exponential growth of trajectories
- Exponential moving averages
- ext4
- extended resource
- External Authorization
- external models
- External Secrets Operator
- externality
- extism
- Extraction attacks
- Extractive QA
- Extractive Summarization
- ExtractNewRecordState
- Extrapolation
- Extrinsic evaluation
F
- f-strings
- F1
- F1 для подграфа
- facet
- faceted search
- facilitator
- fact checking
- fact-checking
- factor graph
- Factual Drift
- factual grounding
- fail-closed
- fail-fast
- fail-safe architecture
- Failed inference
- failed trajectories
- failed trajectory
- Failover threshold
- Failover time
- failure analysis
- Failure Blocking
- failure cases
- failure cost
- failure detection
- Failure mode
- Failure mode: node failure
- failure mode: Ollama не отвечает
- Failure modes
- failure point
- failures_total
- failureThreshold
- Fair share
- Fair use
- fairness metrics
- fairness scheduling
- Fairscale
- Faiss
- FAISS index serialization
- Faiss IVF-PQ
- Faithfulness
- Faithfulness metrics
- faithfulness scorer
- Faithfulness threshold
- FaithScore
- fake device plugin
- fake LLM
- Faker
- fakeredis
- Falco
- Falcon
- Fallback Adapter
- fallback adapters
- Fallback chain
- Fallback message
- fallback model
- Fallback Usage
- fallback-блок
- Fallback-модель
- FallbackContext
- fallocate
- false escalation rate
- False Negative Rate
- False negatives
- false positive
- False positives
- Familiarity bias
- Fan-out/fan-in
- fan_in
- FAPE
- FAQ
- Fast Downward
- fast rejection
- Fast-Conformer
- FastAPI
- FastAPI dependencies
- FastAPI dependency injection
- fastapi-admin
- fastavro
- fastchat
- Faster R-CNN
- faster-whisper
- fastparquet
- fasttext
- FATE
- fatigue
- fatigue bias
- fatigue curve
- Fatigue index
- fault injection
- Faust
- Feast
- feature
- feature definition
- Feature Detection
- Feature engineering
- Feature Engineering for RAG
- Feature flag
- feature flags
- Feature group
- Feature hashing
- Feature monitoring
- Feature selection
- feature store
- Feature validation
- Feature view
- Feature-Aware
- feature-aware draft model
- Feature-Aware Speculative Decoding
- feature-based billing
- feature_usage_logs
- FEC
- federated learning
- feedback embeddings
- Feedback mechanism
- feedback rate
- feedback_log.jsonl
- FEM
- Fencing token
- FEniCS
- fetch_20newsgroups
- Few-shot examples
- Few-shot jailbreak
- Few-shot poisoning
- FewShotChatMessagePromptTemplate
- FFmpeg
- FFN
- FFN dropout
- FFT
- FGSM
- FiD
- Fidelity
- field extraction
- FIFO
- FIFO queue
- Figma
- File System Watcher
- Filebeat
- Filesystem
- Filter selectivity
- filter_order
- filtered ANN
- Filtered ANN Search
- Filtering
- Filters
- Final Answer Match
- final response
- Fine-tune эмбеддера
- Fine-tuned model
- fine-tuning
- Fine-tuning cost
- fine-tuning embedding model
- Fine-tuning LLM for Agents
- Fine-tuning loop
- finish_reason
- FinTabNet
- fio
- Fiqa
- fire-and-forget
- Firecracker
- Firecracker-containerd
- Fireworks AI
- FIRING
- first-class objects
- First-come-first-serve
- first-order optimization
- First-price auction
- first-stage retrieval
- Fisher Information Matrix
- Fivetran
- fixed cost
- Fixed shapes
- Fixed window
- Fixed-size chunking
- fixture
- Flagger
- flake8
- flake8-async
- Flakiness
- Flaky test
- Flaky tests
- flamegraph
- Flamingo
- Flan-T5
- flan-t5-large
- Flan-T5-small
- FlanT5
- Flapping
- FLARE
- Flash Attention 2
- Flash crowd
- Flash Decoding
- FlashAttention
- FlashAttention-3
- FlashDecoding
- flat minima
- Flat planning
- FLAVA
- Fleiss' Kappa
- FlexGen
- Flexibility
- Flickr30k
- Flickr30k Entities
- Flickr8k
- Flink Kubernetes Operator
- flocking
- FlopCountAnalysis
- FLOPs
- flow
- Flower
- FLP theorem
- Fluency
- FLUSHALL
- Flux
- Flywheel
- Focal loss
- Follow-the-sun
- Foolbox
- foraging
- force push
- Forced alignment
- Forced tour
- Formal language
- Formal plan
- formal specifications
- Formal Verification
- formal verifier
- Format Adherence
- format constraints
- format exploitation
- Format prompt
- forward compatibility
- Forward hook
- forward pass
- Foundation models with built-in budget
- four golden signals
- FP16
- FP32
- FP32 master weights
- FP4
- FP8
- FP8 quantization
- FP8 Tensor Core
- FP8-aware training
- FPGA
- FPR@TPR=0.95
- FPS
- frame embeddings
- frame sampling
- Free Tier
- free-riding
- freemium
- freeze
- frequency analysis
- frequency coverage
- frequency penalty
- frequency threshold
- frozen
- Frozen LLM
- FSDP
- fsync
- fsync_after_insert
- full attention
- Full checkpointing
- full compatibility
- Full delegation
- full file strategy
- Full fine-tuning
- Full harness
- full invalidation
- full jitter
- full re-indexing
- full-duplex
- fully-connected network
- FullyShardedDataParallel
- function calling
- Function Permutation
- functional correctness
- Funnel
- Fusing
- Fusion ranking
- Fusion reranking
- Fuyu-8B
- FuyuProcessor
- Fuzzing
- fuzzy matching
- FX graph
G
- G*Power
- G-Pipe
- gain
- Game Days
- GameDay
- Gamification
- GAN
- GAN-based detection
- GAN-style
- Gang scheduling
- Garak
- garbage response
- gated attention
- gated averaging
- gated cross-attention
- gated relevance
- gated residual connections
- Gateway
- gating
- Gating function
- Gatling
- Gauge
- Gaussian Mixture Model
- Gaussian noise
- Gaussian Process
- GC
- GC pause
- GCG
- GCS
- GDPR
- gdrcopy
- GELU
- Gemini
- Gemini 1.5 Flash
- Gemini 1.5 Pro
- Gemini API cache
- GEMM
- GEMM (General Matrix Multiply) в LLM%20%D0%B2%20LLM)
- Gemma
- Gemma-2B
- gen_len
- General agent
- generalization
- Generalized Advantage Estimation
- generate_image
- generation
- Generation confidence
- generation.latency_ms
- Generative attack
- Generative attacks
- generative model
- Generative replay
- generator
- genetic programming
- Geo-routing
- GES algorithm
- GET /prompts/{id}/latest
- GET /tasks/{id
- GGUF
- Ghost Clipping
- GID
- GigaChat
- Gini coefficient
- Giskard
- GIST1M
- Git
- Git Flow
- Git hook
- Git LFS
- Git notes
- Git repository
- git revert
- Git-based approach
- GitHub Actions
- GitHub Actions Summary
- GitHub Copilot
- GitLab CI
- GitOps
- GitPython
- GKE/AKS
- GLaM
- Global + Local Attention
- global attention
- global dictionary
- Global load balancer
- global memory
- global rate limiting
- Global reputation
- GLOO
- GloVe
- GLU
- GLUE
- Gmail API
- GNoME
- Go
- Goal condition rate
- Goal divergence
- Goal Success Rate
- gold documents
- gold standard
- Gold trajectory
- golden examples
- Golden Holdout
- Golden path
- Goldenset
- Goodhart's law
- Google Analytics 4
- Google C4 dataset
- Google Calendar API
- Google Colab
- Google DLP
- Google Generative AI SDK
- Google Pub/Sub
- Google T5X
- Google TPU Pods
- google-api-python-client
- google-auth-httplib2
- google-auth-oauthlib
- Gossip protocol
- GP3
- GPipe
- GPQA
- GPT-2
- GPT-2 Medium
- GPT-2 small
- GPT-2 tokenizer
- GPT-3
- GPT-3.5
- gpt-3.5-turbo
- GPT-4
- GPT-4 eval
- GPT-4 Turbo
- GPT-4o
- GPT-4o mini
- GPT-4V
- GPT2Block
- GPTCache
- GPTQ
- GPU
- GPU acceleration
- GPU affinity
- GPU allocation
- GPU cluster
- GPU Direct
- GPU Direct RDMA
- GPU Inference
- GPU instance
- GPU memory
- GPU memory leak
- GPU memory management
- GPU scheduling
- GPU time
- GPU utilization
- GPU utilization drop
- GPU серверы
- gpu-burn
- gpu-exporter
- GPU-hour
- gpu-memory-utilization
- GPU-экспортёр
- gpustat
- Grace Hopper
- Grace period
- Graceful cancellation
- graceful degradation
- Graceful preemption
- graceful shutdown
- gradcheck
- gradient accumulation
- Gradient Boosted Regression Trees
- gradient clipping
- Gradient compression
- Gradient Conditioning
- gradient descent
- gradient flow
- Gradient inversion attack
- gradient leakage
- gradient masking
- gradient monitoring
- gradient noise
- gradient norms
- Gradient Pulse
- gradient scaling
- Gradient sharding
- gradient step
- gradient synchronization
- gradient-based
- Gradient-based attack
- gradient-based methods
- Gradient-based prompts
- Gradient-based search
- gradients
- Gradio
- gradual fine-tuning
- Gradual Trust
- Grafana
- Grafana Cloud
- Grafana dashboard
- Grafana Tempo
- grammar
- Granger causality test
- Grant
- Graph
- graph breaks
- Graph caching
- graph coloring
- Graph databases for prompt lineage
- graph embedding
- graph imbalance
- Graph instantiation
- Graph Neural Network
- graph optimization
- Graph path
- Graph replay
- Graph-of-Thoughts
- GraphCypherQAChain
- GraphQL
- GraphQL subscriptions
- GraphRAG
- Graphs
- graphviz
- Great Expectations
- greedy mode
- Greedy speculative decoding
- greedy traversal
- green list
- Green ratio
- Gremlin
- grey-box атака
- grid
- grid-level synchronization
- grid_group
- grid_group.sync
- Griffin
- Groma
- Groq
- gross margin
- Gross Profit
- ground truth подграф
- Group Normalization
- Group size
- group strategyproofness
- Group-wise quantization
- Groupcache
- GroupChat
- Grouped-Query Attention
- GroupKFold
- Growth Scenario
- gRPC
- gRPC Load Balancing
- gRPC metadata propagation
- gRPC RESOURCE_EXHAUSTED
- gRPC webhook
- GRPO
- gsarti/synthetic_imdb
- GSM8K
- Guaranteed QoS
- Guard agent
- Guardrails AI
- GUID
- guidance
- Guidance overrides evidence
- Gumbel-Softmax
- gVisor
- Gwet's AC1
- Gymnasium
- gzip compression
H
- H100
- H2D
- H2O
- H3
- half-open state
- hallucinated execution
- hallucination
- Hallucination detection
- Hallucination in reasoning
- Hamming distance
- Hand-crafted jailbreaks
- handcrafted features
- Handlebars
- HandoffSignal
- handover request
- handshake
- Hapax Legomena Ratio
- Happy path
- HAProxy
- Hard constraints
- Hard failure
- hard label
- hard labels
- hard limit
- hard negatives
- hard stop
- Hard watermarking
- Hard-coded Prompt
- Hard-negative mining
- Hardening
- Hardware acceleration
- HarmBench
- harmfulness score
- Harness Engineering
- Harness-engineering
- harness-one
- harness-one/tools
- hash
- hash cache
- hash encoding
- Hash function
- hash index
- Hashing
- hashlib
- hashlib.md5
- Haystack
- Haystack Tracing
- Hazelcast
- HBM
- HBM3
- HCA
- HDBSCAN
- HDR100
- Head-based sampling
- head_dim
- Header Accuracy
- Heading
- health check
- Health check failure
- HEART framework
- heatmap
- Heavy Hitter
- heavy-tailed distribution
- Hebbian learning
- Hedge word penalty
- Helicone
- HellaSwag
- helm
- Helm values
- HELMET
- Helpfulness / Harmlessness
- Helsinki-NLP/opus-mt-en-ru
- Hessian
- heuristics
- hey
- HGX
- Hidden dimension
- hidden representations
- Hidden state
- hidden_size
- Hierarchical
- hierarchical agents
- Hierarchical chunking
- hierarchical context
- Hierarchical delegation
- Hierarchical Hit Rate
- Hierarchical Indexing
- Hierarchical memory
- Hierarchical Planning
- Hierarchical resource quotas
- Hierarchical Retrieval
- hierarchical SLO
- Hierarchical structure
- Hierarchical Summarization
- HierarchicalNodeParser
- hierarchy
- high latency
- high similarity
- high variance
- high-cardinality metrics
- High-level planner
- high-risk context
- High-Throughput
- HighFailureRate
- HighLatency
- highway networks
- HINCRBY
- hinge loss
- HIPAA
- Hiredis
- Histogram
- Histogram binning
- histogram_quantile
- historical pilot
- History
- Hit rate
- Hit rate retrieval
- Hit rate@5
- Hit@3
- hit_count
- HLO
- HMAC
- HNSW
- HNSW+IVF hybrid
- hnswlib
- Hold-out validation
- Holdout set
- Homogeneous data
- Homomorphic Encryption
- Honeypot
- Honeypot запросы
- Hopper GPU
- Hopsworks
- Horizon
- horizontal fusion
- Horizontal Pod Autoscaler
- Horizontal scaling
- Horovod
- host.docker.internal
- hostname
- hot index
- hot key
- hot requests
- hot restart
- hot shard
- hot shard detection
- hot spots
- Hot storage
- hot-reload
- Hot-swap
- hot/warm strategy
- hot/warm индексы
- hotfix
- HotFlip
- HotpotQA
- Hough Lines
- HPC
- HQQ
- HSM
- HTML
- HTML-таблица
- HtmlDiff
- HTN
- htop
- HTR
- HTTP
- HTTP 200
- HTTP 201
- HTTP 404
- HTTP 429
- HTTP 429 Too Many Requests
- HTTP 500
- HTTP 503 Service Unavailable
- HTTP Bulk Insert
- HTTP idempotency
- HTTP PUT
- HTTP Request Node
- HTTP/2 multiplexing
- httpx
- Huber Loss
- HuBERT
- Hugging Face
- Hugging Face CrossEncoder
- Hugging Face Evaluate
- Hugging Face Inference API
- Hugging Face Inference Endpoints
- Hugging Face PEFT
- Hugging Face Trainer
- Hugging Face TRL
- Huggingface CLI
- HuggingFace dataset
- HuggingFace Evaluate
- HuggingFace Optimum
- HuggingFace pipeline
- HuggingFace Transformers
- HuggingFaceEmbeddings
- HuggingFaceH4/ultrachat_200k
- HuggingFaceTB/SmolLM2-360M-Instruct
- Human acceptance rate
- human agreement
- human baseline
- Human evaluation
- Human evaluation costs
- human feedback score
- human judgments
- Human labels
- Human response time
- Human validation
- Human workload
- human-in-the-loop
- HumanEval
- Humanloop
- hybrid approach
- Hybrid architecture
- hybrid CPU/GPU deployment
- Hybrid delegation
- Hybrid detection
- Hybrid eval-set
- Hybrid Learned + HNSW
- hybrid model
- Hybrid scaling
- hybrid scheduling
- Hybrid update strategy
- HybridModel
- HyDE
- Hydra
- Hyena
- Hyena Operator
- hyena-dna
- Hypergraph
- Hypernetwork
- Hyperopt
- Hyperparameter
- hyperparameter search
- hyperparameters
- Hypervolume
- Hypothesis
- hypothesis engine
- Hypothetical
- hypothetical attack
- Hypothetical role
- Hystrix
I
- I/O
- I/O-bound
- IA3
- IAM
- IA³
- ib_core
- ib_read_bw
- ib_umad
- ib_write_bw
- ibdiagnet
- IBM AI Fairness 360
- ibping
- ibroute
- ibsim
- ibstat
- ibstatus
- ibswitches
- ibv_asyncwatch
- ibv_devinfo
- ibv_fork_init
- ibv_poll_cq
- ibv_post_send
- ibv_rc_pingpong
- ibv_reg_mr
- IBV_SEND_INLINE
- IBV_SEND_SIGNALED
- ibv_set_pkey
- ibverbs-utils
- ICA
- Iceberg
- ICI
- Ideal DCG
- idempotency
- idempotency key
- Idempotent consumer
- idempotent increment
- Idempotent upsert
- Idempotent writes
- Identity mapping
- Identity Preference Optimization
- Identity Provider
- IDF
- idle GPU
- if_else
- ignore strategy
- im2col
- Image
- image captioning
- Image patches as tokens
- image retrieval
- Image-grounded Text Generation
- Image-Text Contrastive
- Image-Text Matching
- image-to-image
- image-to-image retrieval
- ImageBind
- imagebind_llm
- Imagen
- ImageNet
- images
- IMAP
- IMDb
- Imitation learning
- imitation model
- Immutable Version
- Imperceptibility
- Implicit feedback
- Implicit KL regularization
- implicit reward
- importance sampling
- importance score
- importance scoring
- Improvement rate
- in-batch negatives
- In-Context Learning
- in-flight embeddings
- In-Memory
- In-memory cache
- in-memory dictionary
- In-memory grid
- In-place rollback
- In-place update
- in-process mock
- Incentive design
- incident.io
- include directive
- increase
- Incremental indexing
- incremental ingestion
- incremental insert
- incremental update
- Independent Draft
- independent draft models
- Independent heads
- INDEX.md
- IndexFlatIP
- IndexFlatL2
- IndexHNSW
- IndexIDMap
- indexing
- IndexIVFPQ
- IndexIVFScalarQuantizer
- IndexScalarQuantizer
- Indirect Prompt Injection
- individual rationality
- inductive biases
- Inductor
- inference
- Inference attack
- Inference cost
- Inference engine
- Inference scheduler
- Inference server
- inference time
- inference-time gradient descent
- inference-time scaling
- inference_mode
- Infini-attention
- InfiniBand
- InfiniBand NDR 400
- InfiniBand partition keys
- infinite loop rate
- Infinity
- Infinity Fabric
- inflight requests
- InfLLM
- InfluxDB
- InfluxDB line protocol
- InfoNCE
- Information Gain
- information gap
- Information loss between agents
- infrastructure cost
- ingestion
- ingestion consumer
- Ingestion latency
- ingestion pipeline
- Ingestion service
- ingestion_error_rate
- initialDelaySeconds
- injection classifier
- Inner Model
- Inpainting
- Input compression
- Input Filter
- Input filtering
- input rails
- Input sanitization
- Insecure Output Handling
- Insecure Plugin Design
- inspect
- Instance Normalization
- instance type
- instruct model
- instruction
- instruction format
- Instruction Formatting
- Instruction prefix
- Instruction tuning
- Instruction-response pair
- InstructLab
- instructor
- instrumentation
- INT4
- INT8
- integral hash
- Integrated Gradients
- Integration test for prompt chain
- Integration testing
- intel-scipy
- intent
- Intent classification
- Inter-agent communication system
- inter-agent messages
- Inter-annotator agreement
- Inter-cluster Distance
- Inter-cluster diversity
- Inter-GPU bandwidth
- inter-judge agreement
- inter-rater reliability
- inter-user variability
- Interactive prototype
- interference
- Interleaved 1F1B
- interleaving
- intermediate layers
- intermediate tokens
- intermediate_answers
- Interpretability
- interruption overhead
- Intersection over Union
- intervention
- interventions
- intfloat/e5-mistral-7b
- intfloat/e5-small-v2
- intfloat/multilingual-e5
- intfloat/multilingual-e5-small
- Intra-cluster Distance
- Intra-cluster diversity
- Intra-list diversity
- Intra-list similarity
- Intra-session diversity
- Intrinsic evaluation
- Intrinsic motivation
- Invalid Prompt
- Invalidation count
- Invariant
- Invariant violation rate
- invariants
- Inverted File Index
- Inverted index
- Inverted list
- IO-aware
- IO-awareness
- IOPS
- IOR
- iostat
- IP-based rate limiting
- iperf3
- iptables
- IQR
- IREE
- IRR
- irrecoverability
- irrelevant text
- IRSA
- ISO 8601
- ISO/IEC 42001
- Isolation Forest
- isolation.level=read_committed
- isolcpus
- isotonic regression
- Istio
- item difficulty distribution
- Item Response Theory
- iterated RLHF
- Iterated Training
- iteration
- iteration-level scheduling
- iterations
- iterative improvement
- Iterative process
- IVF+PQ
- IVFFlat
J
- Jaccard similarity
- Jaeger
- Jaeger connection error
- JaegerExporter
- Jailbreak
- jailbreak chain
- Jailbreak defense
- jailbreak robustness
- jailbreak taxonomy
- Jailbreak-атаки
- JailbreakBench
- JailbreakV-28K
- Jamba
- Janus
- JAX
- Jenkins
- Jensen-Shannon divergence
- Jetson
- JetStream
- jieba
- Jinja2
- Jira
- JIT compilation
- JIT-компилятор
- JIT-компиляция
- jitter
- Jitter buffer
- jiwer
- JMX/MBeans
- JOIN
- joint embedding space
- joint training
- JSON
- JSON logs
- JSON mode
- JSON model
- JSON over HTTP
- JSON schema
- JSON Schema validation
- JSON-LD
- JSON-логгер
- JSONL
- jsonlint
- JTBD
- Judge agent
- JuiceFS
- JUnit
- JUnit XML
- Jupyter Notebook
- JWT Token
K
- K-factor
- K-means
- k1
- k6
- k9s
- k_proj
- Kafdrop
- Kafka
- Kafka compaction
- Kafka Connect
- Kafka Headers
- Kafka lag
- Kafka Lag Exporter
- Kafka Log Cleaner Manager
- Kafka Streams
- Kafka topic
- Kafka transactions
- kafka-python
- Kahneman-Tversky Optimization
- Kaiming initialization
- Kandinsky
- Kata-containers
- kcat
- KD-Tree
- KEDA
- Keep-alive
- Kendall's Tau
- Kendall's τ
- Kerberos
- kernel
- kernel computation
- Kernel density estimation
- Kernel Duration
- kernel fusion
- kernel headers
- Kernel launch
- kernel launch overhead
- kernel trick
- kernels
- Key
- key cache
- key distribution
- key extraction
- Key prefixing
- key words
- Key-value model
- Keycloak
- KeyDB
- KeyError
- keyspace_hits
- keyspace_misses
- KeywordTable
- KG-RAG
- KGW
- Kibana
- KILT
- Kind
- Kirchenbauer watermarking method
- Kirsch-Mitzenmacker
- KL divergence
- KL penalty
- Knapsack problem
- kNN
- knowledge editing
- knowledge graph
- Knowledge Graph from Image
- Knowledge Version
- knowledge_version
- KnowledgeGraphIndex
- known issues
- Kong
- Kosmos-2
- KRaft
- Krippendorff's Alpha
- Krum
- KS-test
- kube-prometheus-stack
- kube-scheduler
- kube-state-metrics
- kubectl
- Kubeflow Pipelines
- Kubernetes
- Kubernetes Admission Controller
- Kubernetes device plugin
- Kubernetes Device Plugin for MIG
- Kubernetes Job
- Kubernetes Jobs
- Kubernetes probe
- Kubernetes Secret
- kubetest
- Kueue
- KV cache compression
- KV cache explosion
- KV cache fragmentation
- KV cache management
- KV cache manager
- KV-cache
- KV-cache compression
- KV-cache replication
- KV-cache reuse
L
- L-Eval
- L1 cache
- L1/L2 cache
- L2 Cache
- L2 distance
- L2 Norm
- L2 Normalization
- L3 cache
- L4
- L7 load balancer
- Label flipping
- Label quality
- label selector
- Label smoothing
- Label Studio
- label_values
- Labelbox
- labeling function
- labels
- LaBSE
- LakeFS
- Lakera Guard
- LAMB
- Lambda Function
- LambdaMART
- LambdaRank
- Lamini
- Lamport clock
- LanceDB
- LangChain
- LangChain AgentExecutor
- LangChain ConversationBufferMemory
- LangChain Hub
- LangChain Red Teaming
- LangChain Tool Calling
- langdetect
- LangFuse
- LangGraph
- LangServe
- LangSmith
- LangSmith Hub
- Language compliance
- Language detection
- language gap
- language representation
- Laplace noise
- Laplace smoothing
- large batch inference
- large batches
- large model
- Last hidden state
- late fusion
- Late interaction
- late-arriving data
- Latency
- Latency costs
- Latency hiding
- Latency injection
- latency overhead
- Latency p50/p95
- latency reduction
- latency requirement
- Latency SLA
- latency SLO
- latency stability
- Latency-based routing
- Latency-Correctness Trade-off Inversion
- Latency-sensitive
- Latent Reasoning
- latent reasoning token
- latent space
- latent space reasoning
- launch overhead
- launch statistics
- LaunchDarkly
- layer splitting
- Layered defense
- LayerNorm
- Layout Analysis
- layout detection
- Layout Optimization
- Layout-Aware Chunking
- Layout-aware parsing
- LayoutLMv3
- LayoutParser
- Lazy creation
- lazy evaluation
- Lazy invalidation
- lazy write
- lazy-loading
- LC-QuAD 2.0
- LD/ST
- LDA
- LDAP
- leader election
- leakage tracking
- leaky bucket
- Leaky ReLU
- Lean
- LeanDojo
- Learnable embeddings
- learnable representations
- Learned Index Structures for ANN
- Learned positional embeddings
- learning curve experiment
- learning from failure
- learning rate
- Learning Rate Schedule
- Learning Rate Scheduling
- learning-to-rank
- Lease
- Least Confidence
- Least connections
- Least critical first
- least-loaded
- lecture search
- left join
- Legal document
- LEGAL-BERT
- Length compliance
- length exploitation
- Length normalization
- Length-based curriculum
- Length-based sampling
- leniency bias
- Lexical diversity
- Lexical gap
- LFAnalysis
- LFU
- LGBMRanker
- libcst
- libibumad
- libibverbs
- library of operations
- librdmacm
- LID
- Lifecycle hooks
- LIFO-эвристика
- LightGBM
- LightLLM
- lightweight BERT
- Lightweight model
- likelihood
- Likelihood ratio
- Likelihood Ratio Attack
- Likert scale
- likwid-perfctr
- LIMA
- LIME
- limited membership
- LimitRange
- Linalg
- Line Chart
- line coverage
- line-based протокол
- Linear
- Linear Artificial Tomography
- Linear attention
- linear complexity
- linear complexity attention
- linear correction
- Linear Decay
- Linear heads
- linear interpolation
- Linear layer
- Linear layers
- Linear Scaling Rule
- Linear SSM
- Linear Transformers
- Linear warmup + linear decay
- Linformer
- link down
- LIPS
- List Preservation
- ListAndWatch
- ListNet
- listwise
- Listwise evaluation
- LiT
- LiteLLM
- LiteLLM Router
- LiveBench
- LiveIdeaBench
- LiveKit
- Liveness probe
- Liveness/readiness probes
- Llama
- Llama 3.1 405B
- Llama Guard
- LLaMA-2-70B
- Llama-3-1B
- Llama-3-70B
- Llama-3-8B
- Llama-3-8B-128k
- Llama-3.1-70B
- llama-cpp-python
- LLaMA-Factory
- llama.cpp
- llama3.2:1b
- LlamaCloud
- LlamaIndex
- LlamaIndex Function Calling
- LlamaParse
- LLaVA
- LLaVA-Bench
- LLC-load-misses
- LLM
- LLM API
- LLM assistant
- LLM augmentation
- LLM calibration
- LLM call
- LLM chain
- LLM compiler
- LLM confidence score
- LLM Cost
- LLM detector
- LLM distillation
- LLM endpoint
- LLM Eval Toolkit
- LLM evaluation
- LLM evaluation metrics
- LLM executor
- LLM fingerprinting
- LLM Gateway
- LLM inference
- LLM inference cluster
- LLM Invoker
- LLM kernels
- LLM logging
- LLM memory
- LLM observability
- LLM pipeline
- LLM price
- LLM production
- LLM server
- LLM streaming
- LLM training
- LLM задачи
- LLM кластер
- LLM с памятью
- LLM-as-a-judge
- LLM-as-firewall
- LLM-as-Judge
- LLM-assessor
- LLM-based detection
- LLM-call classifier
- LLM-firewall
- LLM-generated
- LLM-generated expansion
- LLM-generated hard negative
- LLM-generated hard negatives
- LLM-in-the-loop
- LLM-SR
- LLM-валидатор
- LLM-валидация
- LLM-классификатор
- LLM-оценка риска
- LLM-приложения
- LLM.int8
- LLMChain
- LLMLingua
- LLMOps
- LLMProvider
- LLVM
- LM Contamination
- LM head
- lm-eval-harness
- lm-evaluation-harness
- Lm-format-enforcer
- lm_evaluation_harness
- lmbench
- LMDB
- lmql
- LMSys Chatbot Arena
- Load balancer
- load prediction
- load shedding
- Load testing
- load time
- load_penalty
- loader.py
- Loaders
- Local buffer
- local communication
- local LLM
- Local reputation
- LocalAI
- LocalExecutor
- Locality
- Locality Sensitive Hashing
- LocalQueue
- LocalStack
- LocalStorage
- Locate-Then-Edit
- Lock acquisition time
- Lock contention rate
- Lock falsification
- Lock hold time
- lockfile
- Locking
- Locust
- LOF
- Log Aggregation
- Log Cleaner
- Log Parsing
- log rotation
- log-and-apply
- log-log scale
- log-Mel spectrogram
- Log-probability
- log.cleanup.policy=compact
- log_softmax
- Logger
- Logging levels
- Logical KV-blocks
- Logical Replication
- Logical replication slot
- logistic regression
- Logit Clipping
- logit lens
- Logit masking
- Logit-based fingerprint
- logit-based uncertainty
- logit-манипуляции
- logits
- logits processor
- Logits processors
- LogitsProcessor
- Logprob
- logprobs
- LogQL
- logs
- Logstash
- loguru
- Loki
- Long Context
- Long Context RAG
- Long context reasoning
- long convolutional filters
- long jumps
- Long Range Arena
- Long-context capability
- Long-form
- long-form answers
- long-running agents
- Long-running operation
- long-running задачи
- LongBench
- Longest common prefix
- Longformer
- LongLoRA
- LongNet
- lookahead
- Lookahead decoding
- Loop unrolling
- loose filter
- LoRA
- LoRA merging
- LoRA rank
- lora_alpha
- LoraConfig
- LoReFT
- losetup
- losing response
- Loss
- Loss aversion
- loss landscape
- Loss masking
- loss of diversity
- Loss-based attack
- Loss-based MIA
- Lossless
- lossy side-effects
- Lost in the Middle
- Lost in the Middle prompting
- lost requests
- Low confidence
- low faithfulness
- low latency
- low-bit quantization
- low-confidence highlighting
- Low-level executor
- low-quality filtering
- Low-rank decomposition
- low-rank matrices
- low-rank projection
- Lowering
- LPDDR5X
- LPU
- LRU
- LRU-кэш
- LRU-эвакция
- lru_cache
- LSH attention
- LSTM
- LTV
- Lua filter
- Lua-скрипт
- Lua-скрипты
- Lucene
- Lunary
- Lustre
- lxml
M
- m16n16k16
- m16n64k16
- m16n8k16
- m64n16k16
- m64n64k16
- m8n8k32
- Machine epsilon
- machine unlearning
- Magpie
- Mailbox
- Mailpit
- maintenance window
- majority voting
- Make
- Makefile
- malformed response
- malicious embeddings
- Mamba
- mamba-ssm
- MambaBlock
- MambaFormer
- MAML
- mammoth
- Man-in-the-middle attack
- Mann–Whitney U
- Manual
- manual commit
- manual reprocess
- Manual Review
- manual spans
- many-shot
- map of repo
- mapping
- MapStore/MapLoader
- margin
- Margin Sampling
- MARGIN-режим
- marginal value
- Markdown
- Markdown table
- Market-based delegation
- Marlin kernel
- Marquez
- mask and insertion
- Mask R-CNN
- mask-and-fill
- Masked Image Modeling
- masked language modeling
- Masking loss
- Materialization
- Materials Project
- Materials Project API
- Math
- math reasoning
- Math-500
- math_verify
- Mathlib
- MathQA
- Matplotlib
- matrix factorization
- Matrix multiplication
- Matrix Scaling
- matrix units
- Matryoshka evaluation
- MatterGen
- Matthews Correlation Coefficient
- MATTR
- max attention weight
- max entropy
- max probability
- Max sequence length
- Max similarity to holdout
- max tokens
- max-batch-prefill-tokens
- max-model-len
- max-num-batched-tokens
- max_attempts
- max_batched_tokens
- max_children
- max_degree
- max_delegations
- max_depth
- max_insert_block_size
- max_iterations
- max_length
- max_locked_memory
- max_new_tokens
- max_num_seqs
- max_position_embeddings
- max_retries
- max_seq_length
- max_split_size_mb
- max_steps
- max_tokens
- maximum
- Maximum Calibration Error
- Maximum Mean Discrepancy
- maximum steps exceeded
- maxLength
- maxmemory
- maxmemory-policy
- mbw
- MCP
- MCP Client
- MCP Server
- MCTSAgent
- MCTSNode
- MDLM
- mDNS
- mdtest
- Mean Absolute Error
- Mean pooling
- mechanism design
- Mechanistic interpretability
- Median-based aggregation
- mediasoup
- Medusa-2
- MEGA
- Megablocks
- MegaByte
- Megatron-LM
- Mel-спектрограмма
- Mel-шкала
- Mellanox ConnectX
- membership inference attack
- Memcached
- MemGPT
- MEMIT
- memmap
- memmap_threshold_kb
- Memorization
- Memorization vs. generalization trade-off
- Memory
- Memory & Persistence
- Memory Bandwidth
- memory bandwidth bottleneck
- memory bandwidth utilization
- memory bank
- memory binding
- memory blocks
- memory coalescing
- memory compression
- memory consolidation
- memory corruption
- memory coverage
- memory embeddings
- memory footprint
- memory fragmentation
- memory management
- Memory Networks
- Memory Overhead
- Memory Overhead Ratio
- Memory Pattern
- memory planning
- Memory poisoning
- Memory pool
- Memory profiling
- Memory prompt
- memory reduction
- memory region
- memory savings
- memory stall ratio
- memory stalls
- memory traffic
- Memory Tuning
- memory update
- Memory Updater
- Memory utilization
- memory-bound
- Memory-efficient attention
- Memory-efficient inference
- Memory-optimized ANN
- memory-speed tradeoff
- Mermaid
- Mesa
- MeshOrchestrator
- message bus
- Message dispatcher
- Message Passing Neural Network
- message pipeline
- message replay
- Message Type
- message_id
- messages
- Messages API
- MessageTransport
- meta-evaluation
- meta-learning
- meta-llama/Llama-3.2-3B-Instruct
- Meta-model
- Metadata consistency
- Metadata filtering
- Metadata index
- MetadataReplacementNodePostprocessor
- MetaGPT
- Metal
- METEOR
- metric
- metric drift
- Metric exporter
- Metrics
- metrics counters
- metrics-driven testing
- metrics-server
- MetricsPort
- MHLO
- Micro-interactions
- micro-VM
- microbatches
- Microcopy
- Micrometer
- Microservice architecture
- Microsoft Counterfit
- Microsoft Graph API
- Microsoft TaskWeaver
- MicroTVM
- Middleware
- Middleware Chain
- middleware chains
- Midjourney
- MIG Manager
- MIG profile
- Milestone completion order
- Milestone completion rate
- Milestone evaluation
- Milestone hit rate
- Milvus
- MIME
- MIME type
- Min-max fairness
- Min-Max Scaling
- Min.insync.replicas
- MinHash
- MinHashLSH
- MiniGPT-4
- Minikube
- minimal privileges
- minimum
- MinIO
- MinIO consistency flag
- minLength
- minReplicas
- MIPRO
- MIPROv2
- Mirostat
- MirrorMaker 2
- Mismatch rate
- Missing details
- missing tool
- Mission-Critical Application
- Mistral
- Mistral Large
- Mistral-70B
- Mistral-7B
- Mistral-7B-Instruct
- mitmproxy
- MITRE ATLAS
- MITRE ATT&CK
- mixed batch
- mixed precision training
- mixed-modal
- Mixtral
- Mixtral 8x22B
- Mixture of Experts
- MKL
- ML Certification
- ML Engineer
- ML Model Access
- ML pipeline
- ML workload
- ML-based suggest
- MLaaS
- MLE
- MLflow
- MLflow Tracing
- MLIR
- mlnx-ofed-kernel-dkms
- MLOps
- MLOps pipeline
- MLP layers
- MLP Projection
- MLPerf Inference
- mlx5_core
- MM-Vet
- MMA
- mmap
- MMBench
- MMDiT
- MMHal-Bench
- MMLU
- MNIST
- MNLI
- MobileViT
- mock agent
- Mock API
- mock downstream
- Mock LLM
- Mock message bus
- mock server
- mock-LLM
- mock-воркеры
- mock-провайдер
- Mock-функции
- Mocking LLM
- mocks
- MockTime
- modAL
- Modal window
- model
- Model cards
- model chaining
- Model Compiler
- model depth
- model extraction
- model inversion attack
- Model parallelism
- Model Poisoning
- model ranking
- Model registry
- model selection
- model stealing attack
- Model Theft
- Model unrolling
- Model Updates
- model version
- Model warm-up
- model weights
- Model-based RL
- model.unload
- model_name
- moderation rails
- Module
- Modus ponens
- MOG2
- moment retrieval
- momentum
- MongoDB
- MongoDB Change Streams
- Monitor
- Monitoring and logging
- monitoring delegation
- monitoring errors/latency
- monitoring for LLM applications
- Monitoring stack
- monkeypatch
- monoBERT
- monorepo
- monorepository
- Monotonicity
- Monte Carlo
- Monte Carlo Dropout
- Monte Carlo Tree Search
- moral reasoning attack
- MOS
- Moto
- Mount
- MovedError
- moving average
- MPI
- MPICH
- mpirun
- MPP-движки
- MPS
- MRR
- MRR@10
- MRR@5
- MS MARCO
- MSCCL
- MSCOCO
- MSE
- MT-Bench
- MTBF
- MTEB
- mTLS
- MTTD
- MTTR
- MTU
- multi-agent coordination
- multi-agent debate
- multi-agent jailbreak
- Multi-Agent Orchestration
- multi-agent pipeline
- multi-agent planning
- Multi-agent RAG
- multi-agent system
- multi-agent verification
- Multi-agent workflows
- multi-armed bandits
- multi-context storage
- multi-document question answering
- multi-GPU inference
- Multi-Head Attention
- Multi-hop accuracy
- multi-hop QA
- Multi-hop RAG
- Multi-hop reasoning
- Multi-Instance GPU
- Multi-Latent Attention
- multi-layer DLQ
- multi-layer graph
- multi-modal representation languages
- multi-model
- Multi-model support
- Multi-needle
- multi-objective optimization
- multi-primary
- Multi-Query Attention
- multi-region active-active
- multi-region active-passive
- Multi-region deployment
- multi-region failover
- multi-stage build
- multi-stage retrieval
- multi-step agent
- Multi-step reasoning
- Multi-step research agent
- multi-step retrieval
- multi-step scenario
- Multi-step search
- Multi-Task Optimization
- multi-tenant
- multi-tenant isolation
- Multi-tenant LLM serving
- multi-tenant network
- multi-tenant RAG
- Multi-turn
- Multi-turn attack
- Multi-turn detection
- multi-turn dialogue
- multi-turn QA
- multi-turn scenarios
- multi-turn диалоги
- Multi-vector index
- Multi-vector retrieval
- MULTI/EXEC
- multi_tool
- MultiChainComparison
- Multidimensional IRT
- Multilingual alignment
- multilingual attack
- Multilingual attacks
- Multilingual audio
- Multilingual Retrieval
- multimodal agent
- multimodal embedding
- multimodal encoder
- multimodal LLM
- multimodal retrieval
- multimodality
- Multinomial Diffusion
- Multipart chunk size
- Multipart upload
- multiple annotators
- Multiple Heads
- multiple judges
- Multiple runs
- Multiple sampling
- Multiple Sequence Alignment
- Multiple Testing
- Multiple Testing Correction
- MultipleNegativesRankingLoss
- multiprocessing
- Multitask Learning
- MultiWOZ
- Murmur3Partitioner
- MurmurHash3
- Murphy decomposition
- MusicGen
- Mutating admission
- mutation
- mutual information
- MVP
- MXNet
- mypy
- MySQL
N
- n-gram
- N-gram novelty
- n-gram overlap
- n-grams
- n8n
- Nadam
- Naive RAG
- Namespace
- NaN
- Native function
- Native Protocol
- NATS
- NATS CLI
- Nats-Msg-Id
- nats-py
- natural language
- natural language bottleneck
- Natural Questions
- NCCL
- nccl-tests
- NCCL_BUFFSIZE
- NCCL_DEBUG
- NCCL_IB_DISABLE
- NCCL_IB_HCA
- NCCL_MAX_NCHANNELS
- NCCL_NCHANNELS
- NCCL_NET_GDR_LEVEL
- NCCL_NTHREADS
- NCCL_PROTO
- NCCL_TIMEOUT
- ncu
- NDCG
- NDCG@10
- ndiff
- Near-duplicate
- Needle in a Haystack
- negation
- negative entropy
- Negative Log Likelihood
- negative prompt
- negative prompting
- negative sampling
- negative transfer
- NegotiationRequest
- NegotiationResponse
- NeMo
- NeMo Guardrails
- Neo4j
- NER
- NER model
- Nested cross-validation
- Nested fallback
- Nesterov momentum
- NeSymReS
- network
- network partition
- Network Timeout
- NetworkPolicy
- NetworkX
- networkx path analysis
- neural network
- neural representations
- neurosymbolic integration
- New Relic
- next step accuracy
- next token prediction
- NFKC
- Nginx
- NGINX Ingress
- ngrok
- Nightly tests
- NIST AI 600-1
- NIST AI RMF
- NLI
- NLI model
- nlist
- NLLB
- NLP
- nlpaug
- NLTK
- nltk.word_tokenize
- NLU
- NMT
- nn.Parameter
- nn.Sequential
- nnsight
- No hallucination
- No upfront
- no-answer scenarios
- No-leakage
- no-repeat n-gram size
- no_split_module_classes
- node
- node affinity
- Node Graph
- Node pool
- node selector
- node_exporter
- nodes
- Noise
- noise injection
- Noise Multiplier
- noise-based augmentation
- Noising
- Noisy neighbor problem
- Non-autoregressive inference
- non-autoregressive transformer
- non-blocking
- Non-Maximum Suppression
- Nonce
- NoPE
- NormalFloat4
- normalization
- Normalized edit distance
- Notion API
- Nougat
- Novelty
- novelty effect
- NP-hard
- nprobe
- NPV
- NSFW фильтрация
- nsys
- nsys stats
- NTK-aware RoPE
- NTP
- Null-значения
- Nullable
- num_alloc_retries
- num_checkpoints
- num_heads
- num_workers
- NUMA
- NUMA distance
- numactl
- numastat
- Numba
- Number of deadlocks
- Number of lock retries
- numerical embeddings
- numerical stability
- Numerical Weather Prediction
- numexpr
- numpy
- numpy.mmap
- nvbandwidth
- nvcc
- NVIDIA Container Toolkit
- NVIDIA DCGM Exporter
- NVIDIA GPU Operator
- nvidia-container-toolkit
- nvidia-device-plugin
- nvidia-fabricmanager
- nvidia-peermem
- nvidia-persistenced
- nvidia-smi
- nvidia-smi nvlink -s
- nvidia-smi topo -m
- nvidia-uvm
- NVLink
- NVLink 1.0, 2.0, 3.0
- NVLink 5.0
- NVLink mesh
- NVLink peer access
- NVLink Switch System
- NVLink topology
- NVLink-C2C
- NVMe
- NVMe Offload
- NVML
- nvprof
- NVSwitch 4
- nvtop
- NVTX
- NVTX markers
O
- O(n log n) complexity%20complexity)
- O(n) memory complexity%20memory%20complexity)
- O(n²) calls%20calls)
- O(n²) complexity%20complexity)
- O(n²) memory complexity%20memory%20complexity)
- o_proj
- OAuth
- OAuth 2.0 Client ID
- OAuth-токен
- OAuth2
- OAuth2 Scopes
- obfuscated code
- object detection
- Object store with Git semantics
- object swapping
- Observability Triad
- observation
- Observation of Error
- Obsidian
- occupancy
- occupancy requirements
- Odds Ratio Preference Optimization
- OFED
- off-peak scheduling
- off-policy
- offline batch inference
- offline evaluation
- Offline features
- offline migration
- Offline preference optimization
- Offline RL
- Offline Store
- offline training
- offline-метрики
- Offload
- Offloading
- offset
- Offset management
- OLE
- Ollama
- OMP_NUM_THREADS
- on-call
- on-call rotation
- on-chip memory
- On-demand GPU
- On-Demand Instances
- on-demand price
- on-disk payload
- On-policy
- on_failure_callback
- onboarding
- Onboarding flow
- one-class SVM
- one-hot
- One-time token
- Online auction
- online decoding
- online evaluation
- Online features
- Online fine-tuning
- Online Hard Negative Mining
- online inference
- Online Learned Index
- Online learning
- online reinforcement learning
- Online softmax
- Online Store
- Online vs offline
- online-метрики
- online/offline feature consistency
- ONNX
- ONNX Runtime
- onnxruntime-genai
- Ontology
- OOD encoding
- OOM
- OOV
- Opacus
- Open LLM Leaderboard
- open-ended task evaluation
- Open-weight models
- open_clip
- OpenAI API
- OpenAI Batch API
- OpenAI Embeddings
- OpenAI Evals
- OpenAI Functions
- OpenAI Moderation
- OpenAI Moderation API
- OpenAI Prompt Caching
- OpenAI SDK
- OpenAI Swarm
- OpenAI Triton Inference Server
- OpenAIEmbeddings
- OpenAPI
- OpenAPI specification
- OpenCL
- OpenCLIP
- OpenCV
- OpenLineage
- OpenMetrics
- OpenMined
- OpenModelica
- OpenRouter
- OpenSearch
- OpenSM
- OpenTelemetry
- OpenTelemetry collector
- OpenTelemetry Python SDK
- OpenVINO
- OpenWeatherMap API
- OpenWebText
- Operability
- Operation coverage
- Operational debt
- Operational Excellence
- Operational Intensity
- Operational Readiness Review
- Operational Review
- Operational Reviews
- Operator
- operator optimization
- Operator Satisfaction Score
- OpEx
- OPQ
- ops/sec
- Opsgenie
- OPT
- OPTIC
- Optical Flow
- Optimal checkpointing
- Optimal specification depth
- optimistic locking
- optimizer
- optimizer sharding
- optimizer state
- optimizer step
- optimizers_config
- optional fields
- Optuna
- orchestration
- Orchestration SAGA
- Orchestrator
- Orchestrator pattern
- Orchestrator-Workers
- ORDER BY
- order sensitivity
- ordering
- Originality
- Orleans
- Orthogonal initialization
- Orthogonal Procrustes
- orthogonal transformation
- OTLP
- OTLP exporter
- out of domain
- Out-of-knowledge query
- Out-of-order events
- out_of_scope
- Outbox pattern
- Outcome Reward Model
- Outlier detection
- Outlier score
- outlier-aware scaling
- outliers
- outlines
- output filtering
- Output manipulation
- Output Parser
- output parsers
- output_scores
- over-constraining
- Over-decomposition
- Over-expansion
- over-prompting
- Over-provisioning
- over-pruning
- over-refusal
- over-specification
- Overage
- overconfidence
- Overcorrection
- Overfitting
- Overfitting detection
- overflow
- overhead
- overhead ratio
- Overlap
- Overlay network
- overoptimization
- Overprovisioning
- Overreliance
- Oversampling
- Oversubscription
- overthinking
- OWASP
- OWASP Top 10 for LLM
- OWASP Top 10 for LLM Applications
P
- P&L
- P-tuning v2
- p2pBandwidthLatencyTest
- p50
- P90
- p95
- p99
- p99/p50 ratio
- Pachyderm
- Packet loss
- packing sequences
- Pact
- Pad Tokens
- padded sequences
- PaddleOCR
- PAE
- PAEF
- page cache
- page swapping
- Page-Hinkley
- Paged Attention
- paged optimizer
- Paged Optimizers
- PageRank
- PagerDuty
- PAIR
- Pair representation
- paired t-test
- Pairformer
- pairwise
- pairwise agreement
- Pairwise attention
- Pairwise comparison
- pairwise comparisons
- Pairwise cosine distance
- Pairwise distance
- pairwise embedding distance
- Pairwise loss
- pairwise ranking
- pairwise ranking loss
- PaLM
- PaLM 2
- pandas
- pandas DataFrame
- PandasLFApplier
- Pandera
- parallel branching
- Parallel fallback
- parallel forward pass
- Parallel prefix sum
- Parallel scan
- parallel verification
- parallelism
- parallelizability
- Parallelization
- Parameter-Efficient Fine-Tuning
- Parameterized query
- parameters
- Paraphrasing attack
- Paraphrasing query
- Parent Document Retrieval
- parent span ID
- parent-child chunks
- Parent-child retrieval
- Pareto analysis
- Pareto frontier
- Pareto principle
- Parquet
- Parrot
- Parser
- parsing
- partial data
- partial failure UI
- Partial Harnessing
- Partial hypotheses
- Partial upfront
- partial-response rate
- PartialHarness
- PartialPlan
- Particle Swarm Optimization
- Partition
- partition function
- partition key
- partition tolerance
- Partitioning
- Pass
- Pass Rate
- Pass@1
- Pass@k
- patch
- Patch Embedding
- patch encoder
- Patch match
- PATE
- Path Accuracy
- Path Efficiency
- Path traversal
- Path-level evaluation
- Path-level metrics
- Pathlib
- pattern detection
- pattern matching
- Paxos
- pay-per-token
- pay-per-use
- Payback period
- Payload
- Payload index
- Payload splitting
- payload-индексы
- Payment rule
- PC algorithm
- PCA
- PCIe
- PCIe bottleneck
- PCIe fallback
- PCIe Gen5
- PCIe root
- PCIe switch
- PCIe transfers
- PDDL
- pdfminer.six
- pdfplumber
- pdsh
- Peak memory
- Pearson correlation
- peeking
- Peer-to-Peer
- peer-to-peer bandwidth
- Peer-to-peer delegation
- peer-to-peer interaction
- PeftMixedModel
- PeftModel
- Pegasus
- PendingAction
- Penetration Testing
- PEP 8
- per agent
- Per agent rate limiting
- Per channel rate limiting
- Per priority rate limiting
- Per-agent limit
- per-channel scaling
- per-feature cost breakdown
- per-tensor scaling
- Per-token latency
- Per-token quantization
- Perceived latency
- Percent agreement
- percentile
- perceptual loss
- perf stat
- performance
- Performance Drift
- performance tests
- Performer
- perfquery
- permission_denied
- Permutation
- permutation invariance
- Permutation test
- Perplexity
- perplexity analysis
- perplexity anomaly
- Perplexity change
- Perplexity filtering
- Perplexity gain
- perplexity-based detector
- Persistence
- persona modulation
- Personalization
- Perspective API
- perturbation
- perturbation consistency
- Perturbation Rate
- perturbation-consistency
- pessimistic locking
- Pessimistic Scenario
- Pet-project
- PG-19 dataset
- pg_notify
- pg_partman
- pgcrypto
- PGD
- pgoutput
- pgvector
- Phantom
- phased rollout
- PHB
- PHI
- Phi-2
- Phi-3-mini
- physical attack
- Physical isolation
- Physical KV-blocks
- Pickle
- PII
- PII Detection
- PII leakage
- PII masking
- PII rate
- PII redaction
- pika
- Pilot set
- pin_memory
- Pinecone
- Pinned memory
- pipeline architecture
- Pipeline bubble ratio
- pipeline bubbles
- Pipeline flush
- Pipeline parallelism
- Piper
- piper.cpp
- Pitch Deck
- pivot_root
- PIX
- Pixie
- placeholder
- placeholders
- plain text
- Plan
- Plan Accuracy
- Plan coherence
- Plan Completeness
- Plan Correctness
- Plan deviation score
- Plan Efficiency
- Plan manipulation
- Plan quality
- Plan-and-Execute
- Plan-and-Solve
- PlanAndExecute
- Planner
- Planner/Executor Architecture
- planning
- Planning alignment
- planning model
- Platt scaling
- Playground
- plotly
- Plotly Dash
- PlotQA
- plugins
- plural
- POC
- Pod
- Pod Disruption Budgets
- Pod priority
- pod_count
- Poetry
- Point estimate
- point-in-time
- point-in-time correctness
- Point-in-time recovery
- Point-to-point communication
- pointwise
- pointwise fusion
- Poisson arrival
- policies
- Policy
- Policy as code
- Policy evaluation
- policy gradient
- polling
- Polly
- Pool
- Pooling
- POPE
- Popular POPE
- port-forward
- Porter stemmer
- Portkey
- Position bias
- position bias ratio
- Position Encoding
- Position Interpolation
- Position-aware metrics
- Position-Based Model
- positional invariance
- posix_memalign
- Post-filter
- Post-filtering
- Post-flight check
- Post-hoc Calibration
- post-hoc correction
- post-hoc explanation
- Post-hoc rationalization
- Post-ingestion checks
- post-norm
- Post-processing
- post-processing filter
- Post-processing filters
- Post-retrieval
- Post-training quantization
- PostgreSQL
- PostHog
- Postman
- postmortem
- power analysis
- Power law
- PPOTrainer
- PR
- Pre-baked responses
- pre-commit hook
- Pre-fill
- Pre-filtering
- Pre-flight check
- Pre-ingestion checks
- pre-normalization
- Pre-push hook
- Pre-retrieval
- pre-tokenization
- pre-training
- pre/post conditions
- precision
- Precision exceptions
- precision-recall
- Precision/Recall
- Precision@5
- precision@k
- Precomputed features
- precomputed norms
- preconditions and effects
- predicated execution
- predicated instructions
- predict_linear
- predict_proba
- Predictive scaling
- preemption
- Preemption by recomputation
- Preemption by swap
- preemption overhead
- Prefect
- preference agreement
- preference data collection
- preference distributions
- preference simulation
- Preference tuning
- preferred trajectory
- prefill
- prefill stage
- prefix caching
- prefix hashing
- Prefix injection
- Prefix-tuning
- PrefixSpan
- PReLU
- prepositions
- presence penalty
- Presidio
- PreStop hook
- Pricing model
- pricing per token
- primacy effect
- Primary
- Primary Key
- primary storage
- primary/secondary replication
- Principle of Least Privilege
- prioritization
- Priority
- Priority (Weighted) Routing%20Routing)
- Priority = bid / compute
- Priority ceiling
- priority inheritance
- priority inversion
- priority queuing
- priority-based scheduling
- Privacy Accounting
- Privacy attacks
- privacy by design
- Privilege escalation
- proactive replacement
- probabilistic early recomputation
- probabilistic invalidation
- probabilistic label
- Probabilistic Output
- probabilities
- probability distribution
- probe_duration_seconds
- probe_success
- Probing
- Process
- Process reward model
- Processing time
- Procfs
- Prodigy
- Producer
- producer failure
- Product Manager
- Product Quantization
- Product Quantization (PQ) parameters%20parameters)
- production
- production evaluation
- production incident
- production logs
- production ML system
- production readiness
- ProductQuantizer
- profile
- profiler
- profiling
- Profit margin
- Profitability
- program
- program compilation
- programmatic labeling
- progress bar
- progressive disclosure
- Progressive Neural Networks
- Progressive training
- Projection
- Projection into LLM space
- projection matrix
- Prolog
- Prometheus
- Prometheus + Grafana
- Prometheus Alertmanager
- Prometheus API
- Prometheus Blackbox Exporter
- Prometheus client
- Prometheus scrape interval
- Prometheus TSDB
- Prometheus-2
- prometheus_client
- prompt
- prompt adaptation
- Prompt building
- Prompt chaining
- prompt completion ratio
- prompt composition
- Prompt compression
- Prompt conditioning
- prompt diff
- Prompt Engineer
- Prompt engineering
- Prompt fragility
- prompt hardening
- prompt hash
- Prompt injection
- prompt language
- prompt leakage
- Prompt lifecycle
- prompt lineage
- prompt linting
- Prompt Management
- Prompt manifest
- prompt observability
- prompt regression suite
- Prompt Regression Testing
- prompt rewriting
- prompt rollback
- Prompt Security
- prompt stealing
- Prompt testing strategies
- prompt tokens
- Prompt Tuning
- prompt versioning
- Prompt-based guardrails
- Prompt-tuning
- prompt_hash
- prompt_template_schema.json
- PromptBench
- Promptfoo
- PromptInject
- PromptLayer
- promptlint
- PromptLinter
- prompts engineering
- PromptTemplate
- PromQL
- Promtail
- PrOntoQA
- Proof-of-personhood
- proof-of-work
- propagators
- property-based testing
- property-based tests
- Prophet
- prospect theory
- Protein language modeling
- Protobuf
- Protocol verification
- Prototype
- proven theorems rate
- provider switching
- provisioning
- Proximal Policy Optimization
- proxy API
- Proxy Goal Convergence
- proxy metrics
- proxy reward
- proxy-модель
- Proxy_buffering
- Pruning heads
- pruning search trees
- pseudo-labels
- Pseudo-relevance feedback
- PSI
- psutil
- psychological safety
- psycopg2
- Ptrace
- PTX
- Publisher confirms
- PubSub
- PubTables-1M
- pull-based
- pull-модель
- Pulumi
- purple team
- pushgateway
- pwr
- PWWS
- PXB
- py-spy
- PyArrow
- pybloom_live
- pybreaker
- Pydantic
- Pydantic BaseModel
- PyFlink
- Pygame
- pyirt
- pylint
- pymatgen
- PyMuPDF
- Pyodide
- PyPDF2
- pyproject.toml
- pyrate-limiter
- pyreft
- PyRIT
- PySceneDetect
- PySpark
- PySR
- PySyft
- pytest
- Pytest fixtures for LLM prompts
- pytest-asyncio
- pytest-cov
- pytest-html
- pytest-httpx
- pytest-langchain
- pytest-mock
- pytest-rerunfailures
- pytest-timeout
- pytest-xdist
- Python control flow
- Python SDK
- python-docx
- python-json-logger
- python-pptx
- PyTorch
- PyTorch Geometric
- PyTorch Lightning
- PyTorch Profiler
- PYTORCH_CUDA_ALLOC_CONF
- pytrec-eval
- pyvis
- PyYAML
Q
- Q-Former
- Q-value
- q_proj
- QA
- QA-based evaluation
- QA-based verification
- QAMPARI
- QASA
- Qdrant
- Qdrant Cloud
- Qdrant filter conditions
- qdrant-client
- QdrantClient
- QEMU/KVM
- QK-normalization
- QK^T
- QLoRA
- Qoder
- QoS
- QPS
- QPS per shard
- qrels
- Quadratic bottleneck
- quality
- Quality degradation
- Quality gates
- quality metrics
- quality score
- quality-cost curve
- Quantization
- quantization-aware scaling
- Quantization-aware training
- quantized
- quantized target
- quantized verification
- quantlib
- Quarantine
- Quasar
- Query
- Query Complexity Classifier
- query complexity distribution
- Query embedding
- query expansion
- query latency
- query metrics
- query reformulation
- Query rewriter
- Query routing
- query set
- Query Tokens
- Query-document alignment
- query-positive-negative triplet
- Query/Key/Value vectors
- query_range API
- query_type
- QueryEngine
- Quest
- question distribution
- question generation
- questions
- queue
- Queue length
- queue length monitoring
- Queue Pair
- queue-based escalation architecture
- queue.Queue
- queue_latency
- QuickCheck
- QuIP
- quorum
- Quota
- Qwen 2.5 1.5B
- Qwen-VL
- Qwen2-1.5B
- Qwen2.5 72B
- Qwen2.5-1.5B
- Qwen2.5-1.5B-Instruct
- Qwen2.5-7B
- Qwen2.5-MoE
R
- RabbitMQ
- race condition
- race condition prevention
- RAdam
- radius perception
- Radix tree
- RadixAttention
- Raft
- RAG
- RAG agent
- RAG chains
- RAG Corpus
- RAG evaluation
- RAG indexing
- RAG orchestrator
- RAG pipeline
- RAG poisoning
- RAG-bot
- RAG-префикс
- RAGAS
- RAGEngine
- rainbow teaming
- ramp
- Random
- Random assignment
- random deletion
- random embeddings
- random features
- Random Forest
- random graph
- Random injection
- random insertion
- Random POPE
- Random projections
- Random Search
- Random seed sampling
- Random swap
- random token drop
- random walk
- Randomisation промптов
- Randomized Smoothing
- Rank-Based Normalization
- Rank-one update
- rank_bm25
- ranking
- ranking improvement
- RankNet
- ranx
- RAPTOR
- rare classes
- rare languages
- rare queries
- Rare Tokens
- rare trajectories
- RASA
- rate
- rate limiting
- rate limits
- rate query
- RateLimitExceeded
- Rating
- raw trajectory
- Ray
- Ray Serve
- razdel
- RC-соединение
- RCA
- RCCL
- RDB
- RDB preamble
- RDF
- RDMA
- RDMA Read
- rdma-core
- rdma_rxe
- RDS
- RDTSC
- Re-planning
- Re-prompting
- Re-ranker
- Re-reading policy
- ReAct Agent
- ReAct prompt
- Reactive scaling
- read replicas
- read-after-write
- Read-after-write consistency
- read-only filesystem
- read-only index
- read-only mode
- read-only rootfs
- readiness delayed
- Readiness probe
- README.md
- ReadOnlyRootFilesystem
- real data
- real data mixing
- real-time factor
- Real-time ingestion
- real-time monitoring
- real-time RAG
- Real-time video understanding
- Real-time voice agent
- Real-time обработка документов
- Real-time признаки из Kafka
- REALM
- Reasoning
- reasoning degradation
- Reasoning depth
- Reasoning errors
- reasoning models
- reasoning schema
- reasoning steps
- Reasoning via Planning
- Recalibration
- Recall
- Recall exceptions
- recall@1
- Recall@100
- Recall@5
- Recall@k
- Recency
- recency effect
- Receptance
- receptive field
- Reciprocal Rank
- recompilation overhead
- recomputation
- Recomputation-based preemption
- Reconciliation
- Reconnection strategy
- reconstruction error
- Record
- record_shapes
- RecordException
- Recording
- recording rules
- Recovery actions
- Recovery rate
- recovery time
- recurrence
- recurrent backpropagation
- Recurrent Block
- Recurrent Depth
- Recurrent GPT
- recurrent memory
- Recurrent Memory Transformer
- Recurrent operation
- Recurrent vs parallel computation
- Recursive
- recursive reduction
- RecursiveCharacterTextSplitter
- red list
- RED metrics
- red teaming
- red teaming certification
- red teaming evaluation
- red teaming loop
- Redis
- Redis Cluster
- Redis Enterprise CRDB
- Redis INFO
- Redis Keyspace notifications
- Redis KV-cache
- Redis List
- Redis Lock
- Redis pipeline
- Redis PubSub
- Redis Queue
- Redis Redlock
- Redis replication
- REDIS SCAN
- Redis Sentinel
- Redis Sets
- Redis Stack
- Redis Streams
- Redis-based rate limiter
- redis-benchmark
- redis-cell
- redis-cli
- redis-py
- redis_exporter
- RedisCluster
- RediSearch
- Redpanda Schema Registry
- Redrive Policy
- reduce
- reduce-scatter
- Reducer
- ReduceScatter
- reduction fusion
- Reference architecture
- Reference point
- reference policy
- reference модели
- Reference-based attack
- reflection
- reflection loops
- reflection module
- Reflexion
- Reformer
- refresh interval
- refreshInterval
- ReFT
- refusal
- refusal hacking
- Refusal on OOD
- refusal rate
- refusal suppression
- Refusal testing
- Refusal to answer
- regex
- Regex-фильтры
- region affinity
- Regions
- register pressure
- Registers
- Registry
- Registry service
- Rego
- Regression rate
- regression threshold
- regret
- Regularization
- regularization retrieval
- rehearsal
- reindex
- REINFORCE
- Reinforcement Learning
- Reinforcement Learning for Index Tuning
- Reinforcement Learning from Human Feedback
- Reinforcement Learning with Explanation Reward
- rejection sampling
- rejection tokens
- relative degradation
- relative improvement
- relative order
- Relative Position Encoding
- Relay
- Relay bus
- Relay IR
- RelayCaching
- Relevance check
- relevance score
- relevance signal
- reliability
- Reliability diagram
- Reliability Engineering
- Reloader
- ReLU
- ReLU attention
- Remote Key
- Remote Procedure Call
- Rendezvous hashing
- repair_rate
- Repeat rate
- repeated error
- repetition penalty
- ReplacingMergeTree
- replay buffer
- Replay реальных диалогов
- replica
- replicas
- Replicate API
- Replication factor
- Replication lag
- reply_to queue
- RepoCoder
- representation engineering
- Representation Level
- representation levels
- reprocess strategy
- Reptile
- reputation decay
- Reputation Score
- Reputation scores
- reputation system
- request batching
- Request classification
- Request Count
- Request ID
- Request-level scoping
- request-response
- request_rate
- requests
- required checks
- required field
- required_variables
- requirements.txt
- reranking
- Resampler
- resampling
- Reserve price
- reserved field
- Reserved GPU
- reset timeout
- resharding
- residual blocks
- residual connection
- residual connections
- Residual dropout
- residual stream
- Residual Vector Quantization
- residual vectors
- Resilience4j
- ResNet
- ResNet-18
- ResNet-50
- Resolution rate
- resource cleanup
- resource limits
- resource manager
- ResourceFlavor
- ResourceQuota
- Response Consistency
- response safety
- response_quality_score
- responses
- respx
- REST
- Result validation
- result verification
- retention policy
- Retention Rate
- RetinaNet
- retrieval
- Retrieval agent
- retrieval context
- retrieval degradation
- retrieval latency
- retrieval logs
- Retrieval metrics
- retrieval miss
- retrieval pipeline
- Retrieval Quality
- retrieval results distribution
- retrieval strategy
- Retrieval success rate
- retrieval-based hard negative
- retrieval-based hard negatives
- Retrieval-Generation Correlation
- retrieval.latency_ms
- RetrievalQA
- retrieved_chunks
- retrospective
- Retrospective analysis
- retry
- Retry count
- retry rate
- Retry storm
- retry storm mitigation
- Retry Topic
- Retry with deduplication
- Retry with exponential backoff
- retry_delay
- retryable / non-retryable
- Revenue
- revenue per request
- Revenue Streams
- Revenue Structure
- Reverse Instruction
- reverse proxy
- Review
- revision
- Reward
- reward correlation
- reward delay
- reward hacking
- reward model
- Reward Normalization
- Reward Scaling
- Reward score
- Reward shaping
- Rewrite prompt
- Rewrite-Retrieve-Read
- RGB
- RICE framework
- Ring
- Ring all-reduce
- ring attention
- Ring Attention with Load Balancing
- risk assessment
- risk score
- RiskEx
- Riva
- RL update
- RL4LMs
- RLAIF
- RLHF Evaluation Suite
- RLlib
- RMSD
- RMSE
- RMSNorm
- RMSProp
- RNN
- RoBERTa
- robust aggregation
- Robust child
- robust training
- robustness
- Robustness Evaluation
- Robustness Gym
- Robustness Score
- robustness to overfitting
- Robustness@k
- ROC curve
- ROC-AUC
- RoCE
- ROCE v2
- RocksDB
- ROCm
- ROCProfiler
- ROI
- Role
- role differentiation
- role prompting
- Role-based
- Role-based decomposition
- role-play
- Role-play / persona
- Role-play attack
- role-play jailbreak
- RoleBinding
- Roleplay jailbreak
- Roles
- rollback
- rollback delegation
- Rollback frequency
- Rollback loop
- rolling baseline centroid
- Rolling Buffer Cache
- rolling cache
- Rolling deployment
- rolling restart
- Rolling update
- Rollout policy
- rollouts
- ROME
- RonDB
- roofline model
- Root span
- RoPE
- rotation matrix
- ROUGE
- rouge-score
- Round-robin
- Route53
- Router
- Router Collapse
- Router LLM
- router model
- Router prompt
- routing
- Routing entropy
- row-based retrieval
- Row-level locking
- Row-wise
- rowmax
- RPC queue
- RPO
- RRF
- RSS Feed
- RTO
- RTSP
- RTT
- RTX 4090
- ru_core_news_lg
- ruBERT
- RuBERT NLI
- RuBERT-score
- rubric
- rubric-based evaluation
- ruff
- ruGPT-3.5
- rule-based checks
- Rule-based classifier
- Rule-based executor
- Rule-based filtering
- rule-based reward
- rule-based reward model
- Rule-based routing
- Rule-based suggest
- rule-based validation
- rule-based системы
- rule_files
- RULER
- rules
- Run-time verification
- run_id
- runaway costs
- runbook
- Running queue
- runtime
- Runtime detection
- runtime tracing
- Runtime validation
- RuntimeError
- RuShareGPT
- Russian SuperGLUE
- RuTurboAlpaca
- RWKV
- Rényi DP
S
- S3
- S3 consistency
- S3 events
- S3 Glacier
- S3 timeout
- S3 Versioning
- s3fs
- S4
- S5
- SaaS
- sacrebleu
- safari
- Safe retries
- Safetensors
- safety
- Safety & Guardrails
- safety alignment
- safety benchmarks
- safety case
- safety filter
- Safety fine-tuning
- Safety Valve
- safety valves
- safety-utility trade-off
- Safety/security
- SafetyBench
- SAGA pattern
- SageMaker
- SageMaker Batch Transform
- Saiga
- saliency maps
- Salient weights
- Sama
- Sample Efficiency
- sample ratio mismatch
- sample size
- sample size determination
- Sampler
- samples
- sampling
- sampling probability
- sandbox escape
- sandwich technique
- sanitizer
- Sanitizing parsing
- Sanity check
- SAT
- SAT-решатель
- saturation analysis
- saturation gap
- Saturation point
- save time
- Savings Plans
- SBOM
- scalability
- scalar product
- Scalar quantization
- scalar rating
- Scale
- Scale AI
- scale-and-add
- Scale-to-zero
- scale-up/down
- Scaled dot-product attention
- ScaledObject
- scaling factors
- Scaling Laws
- SCAN
- ScaNN
- scapy
- Scatter-Gather Element
- Scenario
- scenario attack
- Scenario-based routing
- scene detection
- Scene Graph
- Scene Graph Generation
- SCF
- schedule-based scaling
- schedule_interval
- scheduled retraining
- Scheduled RI
- scheduler extender
- Scheduler policy
- Schema Compatibility
- Schema compliance
- schema drift
- schema evolution
- Schema registry
- schema resolution
- schema validation
- Schema-Activated In-Context Learning
- schema-valid data
- schemaless
- Scientific formalization
- scikit-activeml
- scikit-learn
- scikit-optimize
- SciPy
- scipy.integrate.solve_ivp
- scipy.optimize.minimize_scalar
- scipy.spatial.distance
- scipy.stats
- scipy.stats.entropy
- Scissorhands
- Scope
- Score normalization
- score_threshold
- Scorer
- Scorers
- scoring
- Scoring rubric
- scrape
- scrape interval
- scrape_config
- scrape_configs
- scrape_interval
- Scratch
- SCROLLS
- Scrubbing
- SDXL
- Seaborn
- Sealed Secrets
- Search engineering
- seasonality
- SecAgg
- SecAgg+
- Seccomp
- second opinion
- Second-price auction
- Secondary
- secret rotation
- Secret sharing
- Secrets Store CSI Driver
- SecretStore
- Section Recall@k
- Secure Aggregation
- secure containers
- Secure Multi-Party Computation
- seed
- seed examples
- Seed pool
- seed-факты
- seed.py
- Segment caching
- segments
- Seldon
- Seldon Core
- SELECT ... FOR UPDATE
- Selection
- selective activation recomputation
- Selective Attention
- Selective checkpointing
- Selective Context
- Selective memory
- selective pruning
- Selective scan
- Selective shedding
- Selective state space
- selectivity
- Self-Ask
- self-BLEU
- self-chat
- Self-contained query
- self-correcting LLMs
- self-correction
- self-correction loop
- Self-critique through pairwise
- Self-Debugging
- self-diagnosis
- Self-enhancement bias
- self-healing
- self-healing pipeline
- self-hosted
- Self-hosted LLM
- Self-hosted models
- Self-improvement
- self-improvement loop
- Self-instruct
- self-judge
- self-organization
- Self-paced Learning
- self-play
- Self-QA
- Self-RAG
- Self-reflection
- self-reported incidents
- Self-schema generation
- Self-Speculative Decoding
- Self-Supervised Loss
- self-supervised tool use
- self-supervision
- self-training
- SelfCheckGPT
- semantic cache
- Semantic Caching
- Semantic chunking
- Semantic coherence
- Semantic Coherence Score
- semantic comparison
- semantic compression
- semantic conventions
- Semantic distance
- Semantic diversity
- semantic drift
- Semantic duplicate
- semantic entropy
- Semantic function
- Semantic gap
- semantic HTML
- Semantic idempotency
- Semantic Kernel
- Semantic loop detection
- semantic ranking
- Semantic similarity check
- semantic tag
- Semantic Versioning
- semantic watermark
- SemanticChunker
- semaphore
- SendHandle
- sensitive data
- Sensitive Info Disclosure
- sensitivity analysis
- sensitivity curve
- Sensitivity Table
- Sensor
- sent_tokenize
- sentence embeddings
- Sentence-level attack
- sentence-level confidence
- sentence-level evaluation
- Sentence-level NLI
- sentence-transformers
- sentence-transformers/all-MiniLM-L6-v2
- SentencePiece
- SentenceTransformers
- SENTINEL Tokens
- Sentry
- Separate indices
- Separator
- sequence
- sequence alignment
- sequence classification
- sequence graph analysis
- Sequence matching
- Sequence mining
- sequence mode
- Sequence number
- Sequence numbers
- Sequence of steps
- sequence parallelism
- sequence slots
- Sequence-level confidence
- Sequential chain
- sequential delegation
- sequential testing
- serialization
- SerpAPI
- Server-Sent Events
- serverless
- Serverless compute
- Service
- Service Account
- Service Graph
- service mesh
- service name
- ServiceAccount
- ServiceMonitor
- Serving API
- serving framework
- Serving infrastructure
- session
- session history
- Session Management
- session memory
- Session Middleware
- session replay
- session state
- Session Store
- Session window
- Session Worker
- Session-level scoping
- session.timeout.ms
- session_id
- Setex
- Setnx
- SETP
- SetRank
- severity
- severity classification
- SFT
- SFT Model
- SFTTrainer
- SGD
- SGDClassifier
- SGLang
- SHA-256
- Shadow mode
- shadow model
- Shadow testing
- shadow traffic
- Shadowing
- SHAP
- Shape specialization
- shaped reward
- Shapiro-Wilk
- shard key
- shard utilization
- Sharded cache
- sharding
- ShardingStrategy
- Shared context protocol
- shared layers
- Shared plan graph
- Shared prefix
- shared prefixes
- shared state
- Shared Tokenizer
- ShareGPT / OpenAssistant / Dolly
- sharp minima
- sharpening
- Shell
- shell access
- Shikra
- Shingle
- shortcuts
- shuffle instructions
- Side channel
- Side output
- Sidecar
- sidecar pattern
- Siege
- SIEM
- SIFT1M
- SIGKILL
- SigLIP
- Sigmoid
- sigmoid loss
- signal
- Signal alarm
- Signature
- SignatureOptimizer
- SIGTERM
- silhouette score
- SimCTG
- SIMD
- simhash
- Similarity search
- Simple Preference Optimization
- SimpleSpanProcessor
- Simpson's paradox
- Simpy
- SIMT
- simulation
- simulation mode
- Simulation testing
- simulation-based verification
- simulator
- Simulink
- Single representation
- Single Responsibility Principle
- single-stage autoregressive transformer
- single_tool
- Singleflight
- sink ratio
- sink tokens
- Sinusoidal encoding
- Sinusoidal Positional Encoding
- SipHash
- size penalty
- Skeletonization
- skew
- skill
- Skill adoption rate
- skill library
- Skill success rate
- Skills
- skip-grams
- sklearn.metrics
- sklearn.metrics.ndcg_score
- Skweak
- SLA
- SLA compliance
- Slack
- Slack API
- Slack Block Kit
- Slack Bot Token
- Slack Events API
- Slack webhook
- slack-sdk
- Sleep window
- SLERP
- SLI
- Sliding window cache
- Sliding window chunking
- SLO
- SLO violation rate
- SLO-driven
- slot memory
- slot migration
- slot-filling
- Slots
- slowapi
- slowlog
- slowlog-log-slower-than
- SLURM
- SM
- SM occupancy
- Small Initialization
- small LLM
- Small world networks
- Smoke Test
- smoke tests
- smolagents
- smooth quantization
- SMTP
- smtplib
- SnapKV
- snapshot
- Snapshot isolation
- snapshot mode
- SNLI
- Snorkel
- SOAR
- social choice
- Social choice aggregation
- social scoring
- Social welfare
- Soft constraints
- soft label
- Soft labels
- soft limit
- soft TTL
- Soft watermarking
- Soft-embedding
- Soft-label
- Soft-RoCE
- softiwarp
- Softmax
- softmax attention
- Softmax Overflow
- Softmax saturation
- SoftRoCE
- sops
- SoundStream
- source
- Source ID
- Source Verification
- Source weight
- source whitelist
- spaCy
- SPADE
- Span
- Span attributes
- Span Masking
- Span status
- SPANN
- Spanner
- Spark
- Spark Structured Streaming
- SparkSubmitOperator
- SPARQL
- SPARQLWrapper
- sparse attention
- Sparse Autoencoders
- Sparse computation
- Sparse Embedding
- sparse features
- Sparse file
- sparse gradients
- sparse matrix
- sparse MoE
- sparse reward
- Sparse rewards
- sparse softmax
- Sparse Transformers
- spatial hashing
- speaker
- Speaker Diarization
- Spearman correlation
- SPEC.md
- SpecAugment
- Specificity
- speculative decoding
- speculative execution
- speedup
- Spell correction
- Spell-checker
- Sphinx
- SPICE
- Spider
- spilling
- SPIN
- Spin the wheel
- Spinnaker
- spinner
- SPLADE
- Split
- split-brain
- Splunk
- Spoofing attack
- Spot Fleet
- Spot GPU
- Spot Instances
- spot price
- Spot termination
- spot termination notice
- spot termination rate
- Spring Cloud Contract
- spurious correlations
- SQL
- SQL schema
- SQL-инъекция
- SQLAlchemy
- SQLAlchemy 2.0
- SQLAlchemy async sessions
- SQLDatabaseChain
- SQLite
- SQLTableNodeMapping
- SQS
- SQuAD
- SQuAD 2.0
- SRE
- SROIE
- SSD
- sse-starlette
- SST-2
- stability
- Stability AI API
- StabilizationWindowSeconds
- Stable Diffusion
- Stable Diffusion 3.5
- stable version
- Stable-Baselines3
- StableHLO
- stacked bar
- staging environment
- stake
- stale data
- stale-while-revalidate
- Stan
- standard deviation
- Standard RI
- StandardScaler
- STaR
- Startup probe
- Starvation
- state
- State Bloat
- state coverage
- State graph
- State Graphs
- State machine
- state management
- State Manager
- state object
- State reconstruction
- State Recovery
- State Schema
- State snapshots
- state space
- State space exploration
- State Space Model
- State store
- state summarization
- state transfer
- State verification
- state-action-next state
- Stateful
- Stateful testing
- Stateful workflow
- StateGraph
- Stateless
- Stateless RAG
- static analysis
- static batching
- static memory allocation
- Static partitioning
- Static Quantization
- static routing
- static shapes
- stationarity
- statistical distance
- statistical distribution tests
- statistical power
- Statistical tests
- statsmodels
- Steady State
- Step accuracy
- step completion
- step embeddings
- Step Latency
- step merging
- Step Order Accuracy
- Step Success Rate
- step verifier
- step-back prompting
- Step-level supervision
- Step-level training
- step-wise verification
- step_count
- step_number
- stepLR
- steps per session
- sticky assignment
- Sticky sessions
- sticky-сессия
- Stigmergy
- STIX
- STL decomposition
- Stochastic depth
- stochastic rounding
- Stochastic speculative decoding
- Stochasticity
- stop words
- stop_after_attempt
- stop_after_delay
- stop_token
- Storage costs
- storage per shard
- storage system
- StoryBench
- strace
- stragglers
- Stratification
- stratified sampling
- Streaming
- Streaming ASR
- streaming chunking
- streaming data
- Streaming deduplication
- streaming feature pipeline
- Streaming Ingestion
- Streaming parsing
- Streaming pipeline
- streaming tasks
- Streaming TTS
- streaming-агент
- StreamingCallbackHandler
- StreamingLLM
- StreamingResponse
- Streamlit
- StreamReader / StreamWriter
- stress test
- stress-ng
- stride
- Strimzi
- Stripe API
- StripedHyena
- STRIPS
- Strong consistency
- structlog
- Structural consistency
- structural pruning
- Structure preservation
- Structured extraction
- Structured Format
- structured logging
- structured loss metrics
- structured output
- structured output format
- Structured Prompting
- structured representation
- structured representations
- structured response
- Structured table formats
- structuring
- stub database
- Student Agent
- student model
- style bias
- Style Consistency Score
- Subgoal completion rate
- Subgraph
- Subgraph Retrieval Precision
- Subgraph retrieval recall
- Subject
- Subprocess
- Subscription
- subscription_tier
- subshots
- Subtask
- Subtask Completion
- subtle injection
- subtle injections
- subtract max
- subvector
- Success rate
- Successful task completion rate
- SUID
- Summarise prompt
- summarization
- SummarizerMemory
- SummaryIndex
- SuperGLUE
- Supervised autonomy
- supervised loss
- supervisor agent
- Supply Chain
- Supply Chain Vulnerabilities
- Suppress Tokens
- SUPR-Q
- Surge AI
- surrogate objective
- Swap
- swap positions
- Swap-based preemption
- swap-space
- swap-test
- Swapped queue
- swarm coordination
- swarm simulation
- SWE-agent
- SWE-bench
- sweep
- SwiGLU
- Swish
- Switch Transformer
- switching criteria
- swizzle
- Sybil attack
- Sybil protection
- sycophancy
- Symbolic consistency
- symbolic regression
- symbolic representations
- Symlink
- Symmetric quantization
- SymPy
- SyncBatchNorm
- synchronization primitives
- synchronous replication
- synchronous update
- synonym mapping
- Synonym swap
- Synonymizer
- Synthesis
- Synthesizer
- Synthetic batch
- synthetic benchmark generator
- synthetic data collapse
- synthetic data generation
- Synthetic dataset
- synthetic eval collapse
- synthetic eval datasets
- synthetic evaluation
- Synthetic file
- synthetic generation
- Synthetic load
- synthetic request
- syscall interposition
- System cards
- System prompt hardening
- system.query_log
- systolic array
T
- T-lite-instruct
- t-SNE
- t-test
- t3.medium
- T4
- T5
- T5 relative bias
- Table
- Table Extraction Score
- Table format
- Table recovery accuracy
- Table Transformer
- table understanding
- TableFormer
- TableNet
- TableRetrieverQueryEngine
- tablespace
- Tabula
- tabula-py
- tactic
- tag
- Tag-based invalidation
- tags_history.json
- tail latency amplification
- tail risks
- Tail-based sampling
- Tailwind CSS
- tanh
- TAP
- TAPAS
- target CPGA
- target hardware
- Target KL
- target model
- target UP
- target_modules
- targeted attack
- Targeted poisoning
- Task
- task allocation
- Task Completion Rate
- Task curriculum
- Task priority
- task prompt routing
- Task queue
- task taxonomy
- task templates
- task vector
- Task vector arithmetic
- task_id
- TaskGroup
- Tasks per Operator
- taskset
- TaskSpec
- TaskType
- TATR-structure
- Taxonomy
- tc
- TCO
- TCP
- TCP retransmission
- tcpdump
- tctl
- te.LayerNorm
- te.Linear
- Teacher Agent
- Teacher Forcing
- teacher-forcing
- teacher-student
- Team coordination layer
- Tecton
- TEDS
- TEE
- TEI
- Telegram
- telegram bot
- teleprompter
- Temperature
- temperature response
- tempfile
- template
- template circuits
- Template injection prevention
- template versioning
- template-based generation
- Temporal
- temporal bounding
- temporal constraints
- Temporal modeling
- Temporal partitioning
- Temporal PDDL
- Temporal Web UI
- Temporalite
- temporary key unavailability
- tenacity
- tenant_id
- Tenseal
- Tensor Cores
- Tensor parallelism
- tensor-parallel-size
- TensorBoard
- TensorFlow
- TensorFlow Federated
- TensorFlow Privacy
- TensorRT Plugin API
- TensorRT-LLM
- termcolor
- Terminal state
- termination notice
- termTimeoutSeconds
- Terraform
- Tesseract OCR
- Test fixtures
- test generation
- test plan
- Test queries
- test set generation
- Test stand
- Test-Time Compute
- Test-time compute scaling laws
- Test-time iteration
- Test-Time Training
- TestClient
- Testcontainers
- tests
- TestsetGenerator
- text
- Text classification
- Text encoder
- Text repetition
- text-embedding-3-large
- text-embedding-3-small
- text-to-image retrieval
- Text-to-SQL
- TextAttack
- TextFooler
- TextRank
- TF-IDF
- Tfidf + LogisticRegression
- TFLOP/s
- TFLOPS
- TGI
- Thanos
- The Pile
- Theorem of Myerson
- theorem proving
- Theory of Mind
- Thesaurus/WordNet
- think/act/observe
- Thompson sampling
- thop
- Thought-Action-Observation loop
- thrashing
- thread
- thread pool
- thread safety
- thread_block
- threading
- threading.Barrier
- threading.Lock
- threading.Timer
- ThreadPoolExecutor
- threat intelligence
- threat modeling
- threshold
- threshold similarity
- threshold-based filtering
- threshold_early_stop
- Thresholds
- Thrift
- throughput
- Thundering Herd
- Tied embeddings
- tiered SLA
- Tiered storage
- TIES-Merging
- tiktoken
- tiled_partition
- tiling
- time
- time series
- Time to fix
- time to verification
- Time window
- time-series analysis
- time.monotonic
- timeit
- timeline
- timeout
- TimeSformer
- timestamp
- timestamps
- timm
- TinyBERT
- TinyDB
- TinyLlama
- TinyStories
- TIR
- tit-for-tat
- TLA+
- TLS
- TLS 1.3
- TLS/SASL
- TMA
- tmpfs
- Together.ai
- Toil reduction
- token
- Token binding
- token bucket
- Token budgets
- token concatenation
- token cost
- token economics
- Token efficiency
- token leak
- token leakage
- Token manager
- token manipulation
- token masking
- token overshoot
- Token repetition removal
- token smoothing
- Token smuggling
- token usage
- Token-based payment
- Token-level caching
- token-level confidence
- token-level confidence estimation
- Token-level evaluation
- Token-level matching
- token-level representations
- token-level scheduler
- Tokenization of coordinates
- tokenizer
- tokens per second
- tokens per word
- tokens_wasted
- TokenTextSplitter
- TokenTracker
- Tombstone message
- Tombstone record
- Tombstone records
- Tool
- Tool Accuracy
- Tool Call Accuracy
- tool call consistency
- Tool call rate
- Tool correctness
- Tool Degradation with Availability Masking
- Tool Drift
- Tool Executor
- Tool failure
- tool injection
- Tool integration
- tool misuse
- Tool misuse rate
- tool overuse
- Tool prompt
- Tool role
- Tool selection
- tool selection learning
- Tool Success Rate
- Tool System
- tool testing
- Tool Timeout
- Tool trace accuracy
- Tool Usage Accuracy
- tool use accuracy
- Tool use alignment
- Tool Validation
- tool verification
- Tool Versioning
- Tool-level attack
- tool_call_failure
- Toolformer
- ToolValidationError
- Top-1 selection
- top-5-10
- top-k
- top-k KL divergence loss
- Top-k routing
- Top-k sampling
- Top-K sparsification
- Top-p (nucleus) sampling%20sampling)
- Top-token confidence
- topic modeling
- Topics
- topk
- topology
- topology matrix
- topology-aware scheduling
- torch memory stats
- Torch-MLIR
- torch.autograd.Function
- torch.bmm
- torch.compile
- torch.cuda.amp.autocast
- torch.cuda.empty_cache
- torch.cuda.max_memory_allocated
- torch.cuda.memory_snapshot
- torch.cuda.memory_summary
- torch.cuda.set_per_process_memory_fraction
- torch.distributed
- torch.distributed.optim
- torch.jit.script
- torch.no_grad
- torch.utils.checkpoint
- torch.utils.cpp_extension
- TORCH_DISTRIBUTED_DEBUG
- TorchDynamo
- TorchMetrics
- torchrun
- torchvision
- toroidal topology
- Torrance Tests of Creative Thinking
- Total cost per session
- Total Revenue
- toxic content
- Toxicity filter
- Toxicity score
- Toxiproxy
- TPOT
- TPU
- Tqdm
- Trace context
- Trace propagation
- trace validation
- traceability
- traceback
- TraceId
- TraceManager
- traceparent
- TraceQL
- TracerProvider
- traces
- tracestate
- tracking state
- trade-off
- trade-off качество/латенси
- Traefik
- Train set
- Train-serve skew
- train/test split
- training
- training cost proportionality
- Training Data Poisoning
- Training dataset
- training objective
- Training Stability
- TrainingArguments
- trajectories
- trajectory
- trajectory accuracy
- trajectory coverage
- trajectory distillation
- trajectory divergence
- Trajectory Exact Match
- trajectory graph
- trajectory optimization
- Trajectory reward
- Trajectory similarity
- transaction
- Transaction ID
- transactional.id
- transferability
- transform
- Transformer
- transformer block
- Transformer Engine
- Transformer-XL
- transformer_lens
- TransformerBlock
- TransformerLens
- transformers
- transitive closure
- transitive dependencies
- translation
- Translation attack
- Translation role
- transmission overhead
- TransNetV2
- transport layer
- transpose
- traversal-запрос
- Treatment
- TREC
- TREC Robust
- Tree
- Tree Attention
- Tree attention mask
- Tree Cache Management
- tree search
- Tree Search Agents
- tree-based decoding
- trend analysis
- Trigger
- trigger_rollback
- triggers
- Trimmed mean
- Trimming attack
- Triple
- Triplet loss
- TripletDataset
- Triton Inference Server
- TrOCR
- true objective
- TrueSkill
- TruLens
- truncated BPTT
- truncation
- trust calibration
- trust model
- trust score
- trust-weighted averaging
- Trusted documents
- truthful bidding
- Truthful mechanism
- truthfulness
- TruthfulQA
- try-catch
- TSDB
- Tsunami
- TTestIndPower
- TTFT
- TTL
- TTL-словарь
- TTPs
- TTS
- TTT Layer
- Tumbling window
- tuned lens
- Tutel
- Two-person rule
- two-phase commit
- Two-phase indexing
- two-stage training
- two-step confirmation
- type hints
- Type I Error
- Type II Error
- Type-token ratio
- TypedDict
- Typer
- TypeScript
- Typical sampling
- Typo attack
U
- U-Net
- U-shaped curve
- UCB constant C
- UDP
- UI
- uint8
- UltraFeedback
- Ultralytics
- UMAP
- unambiguous semantics
- unanswerable question
- unauthorized tool chain
- Uncertainty quantification
- uncertainty sampling
- uncertainty UI
- Underconfidence
- Underfitting
- underflow
- Underprovisioning
- Undersampling
- undo window
- unembedding
- Unicode
- Unicode homoglyphs
- Unicode replacement character
- UnicodeDecodeError
- unified architecture
- Unified embedding
- Unified embedding space
- unified memory
- Unified retrieval
- Unified schema
- unified_diff
- Uniform control flow
- Uniformity
- Unigram
- Union Type
- unique_paths
- Uniqueness
- unit economy
- Unit test for prompt
- Unit testing
- Unitary/toxic-bert
- unittest
- unittest.mock.patch
- Universal Adversarial Triggers
- Universal Transformer
- Unix Time
- Unleash
- Unnatural Instructions
- Unpacking
- Unsloth
- Unstructured
- unsupervised loss
- untargeted attack
- Untargeted poisoning
- UP-Fall
- Up-training
- Upper Confidence Bound
- UPSERT
- upstream
- Usability testing
- usage
- usage hours
- user adoption
- User bias
- User Browsing Model
- User confirmation
- user engagement
- User feedback
- User flow
- User Modeling
- User persona
- User retention
- user satisfaction
- User Story
- User study
- user-based rate limiting
- user_embedding
- user_id
- user_id хэш
- user_id-based split
- user_template
- user_tenure
- useState
- utility per token
- UUID
- UUID v4
- Uvicorn
- UX
- UX metrics
V
- V100
- v_proj
- VAD
- VAE
- Valid Efficiency Score
- validate_schema.py
- Validating admission
- Validation fail reason distribution
- validation metric
- Validation prompt
- Validation set
- VALSE
- VALSE benchmark
- Value
- Value head
- Value Network
- Vamana
- vanishing gradients
- Vanna.ai
- Variable
- variable cost
- Variable costs
- Variable Renaming
- variable-length sequences
- Variance Estimation
- variance normalization
- variance of accuracy
- variational methods
- Variational Speculative Decoding
- Vault
- Vault Agent Injector
- Vault CSI Provider
- VCR.py
- vector DB poisoning
- vector field
- Vector indexes
- vector score
- vector search
- vector similarity
- Vector stores
- VectorIndex
- VectorStoreRetriever
- VectorStoreRetrieverMemory
- vegeta
- Vendi Score
- Vendor lock-in
- venv
- verbosity
- Verbosity bias
- Vercel
- verifier models
- verifier-guided decoding
- version bump
- Version control
- version negotiation
- versioned agents
- Versioned API
- versioned cache
- versioned documents
- Vertex AI
- Vertex AI Batch Prediction
- Vespa
- veth
- vGPU
- Vickrey-Clarke-Groves auction
- VictoriaMetrics
- Vicuna benchmark
- video group
- video indexing
- video summarization
- VideoCLIP
- VideoCoCa
- VideoMAE
- View
- ViLT
- Violation rate
- Virtual contexts
- virtual nodes
- virtual shards
- Virtual Users
- virtualenv
- VirtualService
- VisDial
- Visibility Timeout
- Vision encoder
- Vision-Language Models
- Visit count
- Visual Embedding
- visual expert modules
- Visual Genome
- Visual grounding accuracy
- visual prompt injection
- Visual Prompt Injection Dataset
- Visualization of computational graphs
- ViT
- ViT-L/14
- Vitis AI
- ViViT
- VL-LLM
- VLLM
- vllm:num_requests_waiting
- VLM
- vocabulary projection
- vocabulary size
- Volatile
- Volcano
- Volume Discount
- Voronoi diagram
- Voting
- VQ-GAN
- VQA
- VQVAE
- VRAM usage
- VS Code
- Vulkan
- vulnerability
- vulnerability disclosure policy
W
- w2v-BERT
- W3C Trace Context
- wait_exponential
- wait_random
- Waiting queue
- WAL
- wal2json
- Wall time
- wall-clock speedup
- warm index
- warm standby
- Warm storage
- warmup steps
- Warp
- warp divergence
- Warp group
- Warp scheduler
- Warp schedulers
- Warp scheduling
- warp stall reasons
- warp-level parallelism
- warp_group
- washout period
- WasmEdge
- Wasmer
- Wasmtime
- Wasserstein distance
- WATCH
- WatchError
- waterfall diagram
- watermark
- Watermark Detector
- watermarking
- WatermarkStrategy.forBoundedOutOfOrderness
- Wav2Vec
- wav2vec 2.0
- Wav2Vec2
- wave beam search
- wave decoder
- Wave Decoding
- weak supervision
- Weaviate
- web search
- WebArena
- WebAssembly
- webhook
- Webhooks
- WebP
- WebRTC
- WebShop
- WebSocket
- Weekly seasonality
- Weight Decay
- weight initialization
- weight optimization
- Weight sharding
- Weight sharing
- Weight tying
- Weight-only quantization
- weighted fusion
- Weighted Kappa
- weighted logistic regression
- weighted recall
- Weighted routers
- Weighted routing
- Weighted Scoring
- Weighted voting
- WeightedRandomSampler
- Weights & Biases
- Weights & Biases Prompts
- WGMMA
- WGMMA instructions
- where clause
- Whisper
- Whisper streaming
- Whisper tokenizer
- whisper.cpp
- WhisperFeatureExtractor
- Whistleblowing
- White-box
- white-box extraction
- white-box jailbreak
- whitelist
- whitelist/blacklist
- Whoosh
- Why3
- WhyLabs
- WhyLogs
- Wikidata
- Wikipedia
- Wikipedia abstracts
- Wikipedia API
- Wikitext
- WikiText-103
- WikiText-2
- Wilcoxon signed-rank test
- Wildcard
- WIMBD
- Win rate
- window + watermark
- windowed processing
- winner prediction accuracy
- winning response
- WinoBias
- Wireframe
- WireMock
- Wireshark
- Wizard-of-Oz
- WizardLM
- WKV
- wmma
- wolframalpha
- Word Error Rate
- Word-level attack
- Word-Patch Alignment
- WordNet
- WordPiece
- Work Request
- worker
- worker_prefetch_multiplier
- workers
- Workflow
- World models
- WORM storage
- worst-case error
- WQE
- Write quorum
- write-behind
- Write-through
- write-through cache
- wrk
- WSL
- Wuerstchen
X
- x-max-priority
- X-RateLimit-*
- X-RateLimit-Limit
- X-RateLimit-Remaining
- X-Tenant-ID
- X-Trace-ID
- XACK
- XAI
- Xavier initialization
- XCom
- xFormers
- XFS
- XGBoost
- xgrammar
- XLA
- XML
- XML-like delimiters
- XML/JSON payloads
- XNNPACK
- xPos
- XREADGROUP
- xxHash
Y
Z
- Z-score
- Z3
- ZAB
- Zamba
- Zapier
- zCDP
- ZeRO
- zero downtime
- Zero init
- Zero point
- ZeRO-3
- zero-copy
- Zero-downtime
- Zero-hit rate
- ZeRO-Infinity
- ZeRO-Offload
- zero-order search
- Zero-shot
- Zero-shot attack
- Zero-shot extrapolation
- Zero-shot generalization
- Zero-shot retrieval
- ZeroSCROLLS
- zigzag effect
- Zipf distribution
- Zipkin
- Zod
- ZooKeeper
- ZSTD
А
- Абстрактный парсер
- Автоматизация сценария
- Автономное делегирование
- авторегрессивное декодирование
- агент в production
- адаптивные алгоритмы
- адаптивный лимит
- активационная разреженность
- акустические токены
- акустическое кодирование
- анализ покрытия
- аннотации о деплоях
- Архитектор агентных систем
- асимметричная репликация кэша
- Асимметричное квантование
- Асинхронная индексация
- асинхронная обработка
- ассоциативный сканер
- Атаки инверсии
- аукцион ресурсов
Б
- батчинг embeddings
- беглость
- бесконечный цикл агента
- бинарное дерево
- блок фиксированного размера
- бутылочное горлышко
- бюджет
В
- Валидация ввода
- векторная БД
- векторная дедупликация
- векторный индекс
- векторный поиск
- Версионирование парсеров
- Версионирование эмбеддингов
- весовой коэффициент
- взвешенная сумма скорингов
- взвешенное среднее
- Внешние API
- выполнимость
Г
- галлюцинации
- галлюцинация мультимодальной модели
- генерация аудио
- гибкость
- гибридные архитектуры
- гибридный retrieval
- гибридный поиск
- градиент кросс-энтропии
- градиентная оптимизация
- граф отношений
- граф решений
- Графовые методы
- Групповая стратификация
Д
- датасеты
- двухступенчатый ретривал
- Двухфазная миграция
- деанонимизация
- декартово произведение
- декларативное описание
- декодирующая голова
- Делегирование человеку
- Дерево саммари
- Детектор PII
- детектор циклов
- детекция водяного знака
- детекция повторяющихся действий
- детерминированное распределение трафика
- детерминированные подмножества
- дивергентное мышление
- дискретизация
- дискретизация аудио
- дискретные токены
- дискриминация задания
- долгосрочная память
- дрейф распределения документов
- дрейф распределения запросов
Е
З
И
- Идемпотентность запросов
- иерархическая кластеризация
- Иерархическая координация
- Иерархическое представление
- изоляция
- изоляция данных
- индексы
- индуктивный bias
- Инжект контекста
- инкрементальные вставки
- инкрементальный расчёт
- Интеграция
- исполнимость
К
- калибровка модели IRT
- кардинальность лейблов
- Каскад моделей
- Каскадная конвертация
- Качество относительно full attention
- кластеризация эмбеддингов
- ключевой кадр
- ключевой поиск
- ключевые кадры
- коллективные коммуникации
- Комбинаторный взрыв
- компилятор DSPy
- коннекторные методы
- консистентность данных
- константная память
- Контекст LLM
- Контекстная маскировка
- контракты между агентами
- контроллер бюджета
- конфигурационный файл
- конфигурация сервера
- корпус документов
- косинусная близость
- Коэффициент автономии
- Коэффициент полезного делегирования
- краткосрочная память
- краудсорсинг с верификацией
- Кэширование запросов
Л
- латентная способность
- логирование
- логирование шагов
- логистическое уравнение
- логическая нестрогость
- Локализация данных
- локальная независимость
М
- маскировка
- Материализация матрицы S
- Матрица действий
- матрица перехода A
- матрица проекции B
- матрица проекции C
- межузловая сеть
- мел-спектрограмма
- мел-спектрограммы
- метаданные
- метрика сложности
- метрики успеха
- микро-бенчмарк
- микросервис
- Минимальные привилегии
- Минимизация данных
- многомерная IRT
- многорукий бандит
- Многостраничные таблицы
- Многошаговая jailbreak-атака
- модель Лотки-Вольтерры
- мониторинг
- Мониторинг безопасности
- мониторинг в production
- мультимодальная изоляция
- мультимодальные возможности
- мультимодальные документы
- мультимодальный RAG
Н
- направленные ребра
- неавторегрессивное декодирование
- недетерминированность
- Несопоставимость пространств
- Низкоранговое пространство
О
- Обезличивание
- Облако
- облачные ресурсы
- обнаружение повторяющихся действий
- обратный поиск
- объединенный эмбеддинг
- Объединённые ячейки
- одноразовый OAuth-токен
- оптимизация промптов
- оракул
- оригинальность
- Оркестрация LLM-приложений
- оценка прогресса
- ошибки
П
- пайплайн
- пайплайн автоматического тестирования
- Пайплайн генерации
- параметризованные тесты
- параметризованный тест
- Паттерн Strategy
- патчи
- Переиндексация
- перекомпиляция
- поведенческие сигналы
- Повёрнутый текст
- политика перемещения данных
- Право на забывание
- прайсинг
- Проблема счастливого пути
- программируемые промпты
- промпт агента
- промпт для парафразирования
- промпты
- Псевдонимизация
Р
- рабочая память
- рабочий набор
- рандомизация
- распределение токенов
- Ребаланс
- ребалансировка партиций
- регистрация промптов
- регрессионное тестирование
- рекуррентное обновление
- репликация кэша
- ротация агентов
- Ручная миграция
С
- сегментация трафика
- селективные фильтры
- семантическая амбигуозность
- семантическая дедупликация
- семантическая кластеризация
- семантическая память
- семантические токены
- Семантический маппинг
- семантическое кодирование
- Сепарабельность
- сериализация datetime
- Сессионные контексты
- Сжатие эмбеддингов
- сигнатуры
- Симметрия
- симуляция отказов
- синтетическая генерация датасетов
- синхронизация кэша
- сложность вопроса
- слои размышления
- Смещение фидбека
- Событийная архитектура
- специальные токены
- специфичные токены
- способность модели
- сравнение ответов
- Среда исполнения
- средняя яркость пикселей
- статистическая значимость
- статистический тест
- стеганография
- Страничная организация памяти
- Структурированные промпты
- Структурные фичи
- субквадратичное внимание
- суммаризация таблицы
- сырая accuracy
Т
- таблица страниц
- текстовый RAG
- текстовый промпт
- телепромпты
- тестирование агентов
- тестовые промпты
- тестовый набор запросов
- тестовый сценарий
- типы узлов
- токенизация изображений
- Токенизация состояния
- токены
- топология GPU
- траектория агента
- Транзакционный консюмер
- Транзакционный продюсер
- трансформер-декодер
- трейсинг
- Тримминг
У
- уведомления
- угадывание
- универсальный формат промпта
- Уникальный индекс
- уравнения Лагранжа
- уравнения Навье-Стокса
- успешная сессия
- утечка данных
- Учёт занятости операторов
Ф
- фармакокинетика
- фильтр
- Фильтр на генерации
- фильтр по tenant_id
- Фитнес-функция
- формальные контракты
- форматы хранения промптов
- Фрагментация
- фрагментация данных
Х
Ц
Ч
Ш
Э
- Эвристика BERTScore
- эквивалентность тестов
- экономика агентов
- экспоненциальное сглаживание
- эмулятор LLM-сервиса
- Эндпоинт
- энергопотребление
- Эскалация человеку
Я
- --privileged
- 1 GPU
- 100k документов
- 152-ФЗ
- 1B LLM
- 1F1B
- 1F1B with interleaving
- 1PL модель
- 202 Accepted
- 2:4 sparsity
- 2PL
- 2PL модель
- 3D parallelism
- 3PL
- 3PL модель
- 4-bit inference
- 4-bit quantization
- 409 Conflict
- 4D-параллелизм
- 4th gen
- 4xx
- 5 почему
- 502 Bad Gateway
- 503 Service Unavailable
- 504 Gateway Timeout
- 5xx
- 70B model
- 7B model
- 8-bit
- 8-bit inference
- 8-bit quantization
- 8GB RAM
- @Codebase
- @harness-one/devkit
- launch_bounds
- __syncthreads
- __transaction_state
- _set_padded_sequence
- _set_sequence_lengths
- ∇-Reasoner