EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
Entity records
4748 records · 1/119
- 001
Compute Campus
A concentrated cluster of computing infrastructure built for large-scale AI training and inference, typically including data cen
Website pending - 002
Business Context
The domain knowledge, process state and data semantics an agent needs to act correctly within an enterprise workflow.
Website pending - 003
Token Gateway
An intermediary layer that centralizes model invocation for authentication, quota, routing and usage metering.
Website pending - 004
Agent Hosting
A platform-managed model for deploying, scaling and maintaining the lifecycle of agent services.
Website pending - 005
Grid expansion
Upgrades to transmission and distribution capacity and interconnection facilities to accommodate new electrical load.
Website pending - 006
Token Volume
The total number of tokens processed during inference, commonly used as a proxy for real-world model adoption and workload.
Website pending - 007
Price per Million Tokens
A pricing unit that expresses model inference cost per million input or output tokens, enabling cross-model cost comparison.
Website pending - 008
Fixed Wireless Access
A networking approach that delivers fixed broadband to homes or businesses over cellular wireless networks.
Website pending - 009
Multi-Agent Framework
A system of multiple collaborating agents that decompose tasks and invoke each other to complete complex end-to-end workflows.
Website pending - 010
Biological Large Language Model
A large model trained on biological sequences such as DNA and proteins to learn the distribution of genetic code and generate no
Website pending - 011
Hardware-software-chip-cloud co-optimization
A joint optimization approach spanning software, device hardware, chips and cloud resources to improve end-to-end performance an
Website pending - 012
System-level AI
An architecture that embeds AI capabilities at the OS layer, letting the system orchestrate models and compute for direct use by
Website pending - 013
Long-Context Training
Training methods that use long documents plus synthetic retrieval-distractor and multi-hop QA data to preserve attention effecti
Website pending - 014
FP8 Mixed-Precision Training
A low-precision training scheme that stores some tensors in 8-bit floating point while keeping other stages at higher precision,
Website pending - 015
Column Chromatography
A purification method that passes samples through packed media to separate target particles by size or affinity differences.
Website pending - 016
Sequence Homology Search
Searching large protein sequence databases for evolutionarily related sequences to infer function and distribution.
Website pending - 017
End-to-End World Model
A modeling paradigm that maps raw sensory input directly to action decisions without hand-engineered intermediate modules such a
Website pending - 018
Weight Editing
Directly modifying a neural network's parameter values to alter behavior, as opposed to updating weights through training or fin
Website pending - 019
Automated Test Generation
Using models to automatically produce executable test scripts and assertions from requirements and case designs, reducing manual
Website pending - 020
Change Impact Analysis
Inferring the blast radius of a code change from diffs and business rules to determine which modules and test scopes need focuse
Website pending - 021
Decision Graph
An orchestration pattern that arranges role nodes and conditional branches in a graph so rules and agents collaborate in depende
Website pending - 022
Header-Based Routing
Forwarding traffic at gateways or load balancers based on method and tool names in headers, without parsing request bodies.
Website pending - 023
Multi Round-Trip Requests
An interaction pattern where a tool call raises input requests with opaque state, and the client re-issues the call after collec
Website pending - 024
Stateless Transport
A transport mode where the server keeps no per-session context between requests; each request carries its own protocol version a
Website pending - 025
F1 Score
The harmonic mean of precision and recall, used to summarize classifier performance under class imbalance.
Website pending - 026
Power Capacity
The供电 scale a data center can secure, measured in MW or GW, acting as the physical ceiling on AI compute expansion.
Website pending - 027
Compute Backlog
The total value of signed but not yet recognized compute contracts, a core metric for neocloud financing capacity and revenue vi
Website pending - 028
Repository-Level Code Understanding
The ability to locate relevant files, trace call dependencies and infer change impact across an entire code repository.
Website pending - 029
Cross-Module Modification
A change pattern where one feature edit must be applied consistently across interdependent modules to avoid inconsistent system
Website pending - 030
Capability Transfer
The practice of moving capabilities a model already possesses onto another model or task, achievable via data, distillation, or
Website pending - 031
Gradient Descent
An optimization method that iteratively updates parameters along the negative gradient of a loss to minimize an objective, centr
Website pending - 032
Code Diff
A technique that presents additions, deletions and modifications between two code versions at line or syntax granularity, underp
Website pending - 033
Distributed Storage System
A storage architecture that spreads data across multiple nodes coordinated by consistency protocols to support high-concurrency
Website pending - 034
Parallel Workspaces
Isolated runtime environments per agent that let multiple development tasks proceed concurrently without interference.
Website pending - 035
Sparse Activation Model
An architecture whose total parameter count far exceeds those activated per inference, typically using routing to cut compute co
Website pending - 036
Open Model License
Legal terms governing use, modification and commercial rights of model weights, ranging from MIT and Apache 2.0 to custom commer
Website pending - 037
Model-Product Co-Design
An engineering approach that evolves model capabilities in lockstep with downstream product requirements, feeding real usage sig
Website pending - 038
Model System Card
A standardized document released with a model describing capability boundaries, evaluation results and safety risk assessments.
Website pending - 039
Long-context Modeling
Techniques enabling models to retain effective attention and information use over very long input sequences, involving positiona
Website pending - 040
Long-Horizon Agentic Session
A full agent run that chains many tool calls and environment interactions across an extended decision sequence within one task.
Website pending