model checkpoint
An AI model checkpoint is a saved version of model parameters and related state from a particular point in training or development.
AI Terminology
Search clear AI terminology definitions, acronyms, related concepts and the reviewed videos that explain them.
Showing 501–520 of 705 terms
Clear filtersmodel compression
Model compression reduces an AI model's storage, memory or compute requirements while trying to preserve useful capability.
model consistency
AI model consistency is the degree to which a model produces reliably similar quality and requirement-following across comparable runs.
model controllability
AI model controllability is the degree to which users can reliably direct a model's behavior through instructions and settings.
model drift
Model drift is a change in an AI model's real-world behavior or performance as models, data, users or operating conditions evolve.
model embedding layer
AI model embedding layerAn AI model embedding layer converts discrete input identifiers such as tokens into numerical vectors the model can process.
model failover
AI model failover reroutes a request to another eligible model when the preferred model is unavailable or unhealthy.
model offloading
AI model offloadingAI model offloading moves selected model weights or runtime state from accelerator memory to system memory or storage.
model parallelism
AI model parallelismAI model parallelism splits one model's computation or parameters across multiple processors so they work together on the same workload.
model pool
An AI model pool is a defined set of models eligible to handle a particular task or class of requests.
model portability
AI model portabilityAI model portability is the ability to move a model and its workload between compatible hardware, runtimes or inference providers.
model post-training
AI model post-trainingAI model post-training adapts a pretrained model using additional examples, preferences, rewards or domain data after its initial large-scale training.
model preview
An AI model preview is limited early access to a model before its final availability, terms or behavior are established.
model proxy
AI model proxyAn AI model proxy is an intermediary service that receives model requests and forwards them to one or more compatible models or providers.
model ranking
AI model ranking orders candidate models for a task using defined performance, cost, latency or preference signals.
model scaling
AI model scaling increases resources such as model size, training data or computation to improve an AI model's measured capabilities.
model serving
AI model serving is the infrastructure and software process that makes a trained model available to handle inference requests reliably and efficiently.
model training
AI model training is the process of adjusting a model's parameters so its outputs better satisfy objectives demonstrated by data, feedback or tasks.
multi-head latent attention
MLAMulti-head latent attention compresses attention key and value representations into a lower-dimensional latent form to reduce memory and inference cost.
multi-joint robot coordination
Multi-joint robot coordination controls several joints together so their combined movement produces a stable and accurate physical action.
Knowledge graph
Explore AI concepts
Follow general terms into specialised sub-terms. Select any node to open its definition.