Collection · 93 entries
AI A–Z
Plain-language expansions of the acronyms that fill AI documentation and research papers.
- Terminology
AGI
Artificial General Intelligence — AI capable of performing any intellectual task human beings can perform.
Beginner - Agentic AI
AGY System
An industry-standard acronym standing for Agentic Intelligence System Architecture within agentic ai.
Beginner - Acronym
AI
AI stands for artificial intelligence: computer systems that perform tasks normally associated with human intelligence.
Beginner - Terminology
ALiBi
Attention with Linear Biases — Positional embedding adding linear penalties based on distance.
Intermediate - Terminology
ANN
Approximate Nearest Neighbor — Search algorithms finding close embedding matches quickly.
Intermediate - Acronym
API (AI context)
An API is the interface an application uses to send prompts to a model and receive outputs, without hosting anything itself.
Beginner - Terminology
ASI
Artificial Superintelligence — Hypothetical AI exceeding human intelligence across every field.
Beginner - Terminology
ASR
Automatic Speech Recognition — Automated computational transcription of speech waveforms.
Beginner - Terminology
AWQ
Activation-aware Weight Quantization — Quantization protecting salient weight channels.
Intermediate - Terminology
BM25
Best Matching 25 — Ranking function used by search engines for sparse keyword retrieval.
Beginner - Multimodal AI
CFG
An industry-standard acronym standing for Classifier-Free Guidance within multimodal ai.
Beginner - Terminology
CoT
Chain-of-Thought — Prompting models to output step-by-step reasoning tokens.
Beginner - Terminology
CV
Computer Vision — AI field analyzing and extracting meaning from visual inputs.
Beginner - Acronym
CV (Computer Vision)
CV stands for computer vision: enabling machines to interpret images and video.
Beginner - Multimodal AI
DAC
An industry-standard acronym standing for Descript Audio Codec within multimodal ai.
Beginner - Multimodal AI
DiT
An industry-standard acronym standing for Diffusion Transformer within multimodal ai.
Beginner - Acronym
DL
DL stands for deep learning: machine learning using neural networks with many layers.
Beginner - Fine-Tuning & Optimization
DoRA
An industry-standard acronym standing for Weight-Decomposed Low-Rank Adaptation within fine-tuning & optimization.
Beginner - Terminology
DPO
Direct Preference Optimization — Fine-tuning language models directly on preference pairs without reward models.
Intermediate - RAG & Vector Search
DPR
An industry-standard acronym standing for Dense Passage Retrieval within rag & vector search.
Beginner - Code Generation
FIM Format
An industry-standard acronym standing for Fill In The Middle Code Generation within code generation.
Beginner - Terminology
FlashAttention
FlashAttention — Fast IO-aware attention algorithm optimizing GPU memory movement.
Intermediate - Hardware & Accelerators
FlashAttn
An industry-standard acronym standing for FlashAttention Architecture within hardware & accelerators.
Beginner - Acronym
GAN
GAN stands for generative adversarial network: a generator and a discriminator trained in competition.
Advanced - Core Architecture
GeLU
An industry-standard acronym standing for Gaussian Error Linear Unit within core architecture.
Beginner - Terminology
GGUF
GPT-Generated Unified Format — Binary file format for single-file local LLM loading.
Beginner - Terminology
GoT
Graph of Thoughts — Combining thoughts into arbitrary graph networks for complex problem solving.
Intermediate - Acronym
GPT
GPT stands for generative pre-trained transformer, describing a decoder-only transformer pretrained on text then adapted for tasks.
Beginner - Terminology
GPTQ
Generalized Post-Training Quantization — One-shot post-training quantization algorithm for LLMs.
Intermediate - Terminology
GQA
Grouped-Query Attention — Shared key/value heads for balanced speed and memory usage.
Intermediate - Alignment & Preference Tuning
GRPO
An industry-standard acronym standing for Group Relative Policy Optimization within alignment & preference tuning.
Beginner - Terminology
HNSW
Hierarchical Navigable Small World — Graph data structure for fast approximate nearest neighbor search.
Intermediate - RAG & Vector Search
HyDE
An industry-standard acronym standing for Hypothetical Document Embeddings within rag & vector search.
Beginner - RAG & Vector Search
IVFPQ
An industry-standard acronym standing for Inverted File Index with Product Quantization within rag & vector search.
Beginner - Terminology
KTO
Kahneman-Tversky Optimization — Preference alignment using binary good/bad feedback signals.
Intermediate - Terminology
KV
Key-Value — Key and value vectors stored in GPU memory during autoregressive generation.
Beginner - Acronym
LLM
LLM stands for large language model: a very large neural network trained on text to predict the next token.
Beginner - Terminology
LLMOps
Large Language Model Operations — Operational workflows for deploying and monitoring LLM apps.
Beginner - Terminology
LMM
Large Multimodal Model — Foundation model understanding text, vision, audio, and video inputs.
Beginner - Terminology
LoRA
Low-Rank Adaptation — Parameter-efficient fine-tuning using low-rank decomposition matrices.
Beginner - Terminology
LPU
Language Processing Unit — Groq's custom hardware architecture designed for fast LLM inference.
Beginner - RAG & Vector Search
LSH Index
An industry-standard acronym standing for Locality-Sensitive Hashing for Vector Search within rag & vector search.
Beginner - Acronym
MCP
MCP stands for Model Context Protocol: an open standard for connecting AI assistants to tools and data sources.
Advanced - AI Infrastructure
MCP Protocol
An industry-standard acronym standing for Model Context Protocol within ai infrastructure.
Beginner - Inference & Serving
MCTS
An industry-standard acronym standing for Monte Carlo Tree Search within inference & serving.
Beginner - AI Safety & Ethics
MechInt
An industry-standard acronym standing for Mechanistic Interpretability within ai safety & ethics.
Beginner - Terminology
MHA
Multi-Head Attention — Classic Transformer attention splitting representation into multiple heads.
Beginner - Acronym
ML
ML stands for machine learning: systems that learn patterns from data instead of following hand-written rules.
Beginner - Terminology
MLOps
Machine Learning Operations — DevOps practices applied to machine learning lifecycles.
Beginner - Acronym
MLOps / LLMOps
MLOps and LLMOps are the operational disciplines for running machine learning and language-model systems in production.
Intermediate - Terminology
MoE
Mixture of Experts — Sparse model architecture routing tokens to specialized expert sub-networks.
Intermediate - Terminology
MQA
Multi-Query Attention — Single key/value head shared across all query heads.
Intermediate - Terminology
NLG
Natural Language Generation — Subfield of NLP producing coherent human language text.
Intermediate - Terminology
NLP
Natural Language Processing — Computational field giving computers the ability to understand text.
Beginner - Terminology
NLU
Natural Language Understanding — Subfield of NLP focusing on semantic intent comprehension.
Intermediate - Terminology
NPU
Neural Processing Unit — Dedicated on-device processor for accelerating AI tasks on mobile/laptops.
Beginner - Terminology
OCR
Optical Character Recognition — Extracting text from images and scanned document pages.
Beginner - Terminology
ORPO
Monolithic Preference Optimization — Reference-model-free alignment combining SFT and preference loss.
Advanced - Inference & Serving
PagedAttn
An industry-standard acronym standing for PagedAttention Memory Allocation within inference & serving.
Beginner - Terminology
PEFT
Parameter-Efficient Fine-Tuning — Techniques adapting models by updating a tiny fraction of weights.
Intermediate - Terminology
PPO
Proximal Policy Optimization — Reinforcement learning algorithm standard for RLHF alignment.
Intermediate - RAG & Vector Search
PQ
An industry-standard acronym standing for Product Quantization within rag & vector search.
Beginner - Alignment & Preference Tuning
PRM
An industry-standard acronym standing for Process Reward Model within alignment & preference tuning.
Beginner - Terminology
QLoRA
Quantized Low-Rank Adaptation — 4-bit model fine-tuning with 16-bit LoRA adapters.
Intermediate - Terminology
RAG
Retrieval-Augmented Generation — Fetching external database context to ground LLM generation.
Beginner - Terminology
ReAct
Reasoning and Acting — Interleaving verbal reasoning traces with tool execution actions.
Beginner - Fine-Tuning & Optimization
ReFT
An industry-standard acronym standing for Representation Fine-Tuning within fine-tuning & optimization.
Beginner - Terminology
RLAIF
Reinforcement Learning from AI Feedback — Aligning LLMs via automated AI preference ratings.
Intermediate - Terminology
RLHF
Reinforcement Learning from Human Feedback — Aligning LLMs via human reward signals.
Intermediate - Core Architecture
RMSNorm
An industry-standard acronym standing for Root Mean Square Normalization within core architecture.
Beginner - Terminology
RoPE
Rotary Position Embedding — Relative positional encoding rotating key/query vectors.
Intermediate - RAG & Vector Search
RRF
An industry-standard acronym standing for Reciprocal Rank Fusion within rag & vector search.
Beginner - AI Safety & Ethics
SAE
An industry-standard acronym standing for Sparse Autoencoder within ai safety & ethics.
Beginner - Terminology
SFT
Supervised Fine-Tuning — Fine-tuning base models on instruction-response pairs.
Beginner - Terminology
SimPO
Simple Preference Optimization — Simpler preference optimization using length-normalized implicit reward.
Advanced - Terminology
SLM
Small Language Model — Compact language models (<5B parameters) optimized for edge deployment.
Beginner - Inference & Serving
SpecDec
An industry-standard acronym standing for Speculative Decoding within inference & serving.
Beginner - RAG & Vector Search
SPLADE
An industry-standard acronym standing for Sparse Vector Embedding Protocol within rag & vector search.
Beginner - Terminology
STT
Speech-to-Text — Transcribing spoken audio into written text output.
Beginner - Terminology
SWA
Sliding Window Attention — Attention restricted to a fixed local window size.
Intermediate - Core Architecture
SwiGLU
An industry-standard acronym standing for Swish Gated Linear Unit within core architecture.
Beginner - Agentic AI
TF
An industry-standard acronym standing for Toolformer Action Protocol within agentic ai.
Beginner - Terminology
TF-IDF
Term Frequency-Inverse Document Frequency — Statistical measure reflecting word importance in documents.
Beginner - Terminology
ToT
Tree of Thoughts — Exploring tree-structured reasoning paths with lookahead and backtracking.
Intermediate - Inference & Serving
TPS
An industry-standard acronym standing for Tokens Per Second within inference & serving.
Beginner - Terminology
TPU
Tensor Processing Unit — Google's custom ASIC hardware designed for deep learning.
Beginner - Inference & Serving
TTFT
An industry-standard acronym standing for Time To First Token within inference & serving.
Beginner - Terminology
TTS
Text-to-Speech — Synthesizing natural human audio speech from written text.
Beginner - Multimodal AI
ViT
An industry-standard acronym standing for Vision Transformer within multimodal ai.
Beginner - Terminology
VLM
Vision-Language Model — Neural network processing both visual images and text prompts.
Beginner - Multimodal AI
VQ-GAN
An industry-standard acronym standing for Vector Quantized Generative Adversarial Network within multimodal ai.
Beginner - Multimodal AI
VQ-VAE
An industry-standard acronym standing for Vector Quantized Variational Autoencoder within multimodal ai.
Beginner - Acronym
XAI
XAI stands for explainable AI: methods that make individual model decisions understandable to affected people.
Intermediate