Technical Glossary

Storage driver combining distinct read-only image layers and a read-write layer into a single unified directory tree.

Storage optimization strategy deferring file copies until modifications occur, sharing underlying layer blocks.

Dockerfile design pattern using multiple FROM instructions to separate build-time SDK dependencies from slim runtime artifacts.

Continuous control loop pattern comparing actual cluster state against declared target state in etcd.

Neural network architecture relying entirely on self-attention mechanisms to compute contextual representations in parallel.

Architecture pattern augmenting LLM prompts with relevant context retrieved dynamically from external vector databases.

Parameter-efficient fine-tuning technique freezing base weights and inserting trainable low-rank decomposition matrices.

Alignment technique tuning LLM outputs using reward models trained on human pairwise preference rankings.

Agent prompting framework interleaving explicit step-by-step reasoning thoughts with action tool calls and observation inputs.

Control pattern decomposing high-level user tasks into DAG sub-goals executed sequentially or in parallel by subagents.

Agent self-correction loop evaluating tool outputs or generated code against criteria before returning final output.

Architecture connecting multiple specialized subagents via a central router or swarm topology to solve complex tasks.