implementation detail · filed under evaluation
Isolated container per rollout
Each agent rollout runs in an isolated container built from the task's official image.
Also called isolated container per agent rollout.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
Each rollout runs in an isolated container built from the task's official image, and is capped at 600 agent turns with a 10-hour timeout.
usedunclearin GLM-5.3Z.ai
Filed alongside
Other methods under evaluation :: evaluation harness.
DeepSeek Harness Minimal modeHarborNeMo Evaluator SDKAtomic binary rubric evaluationAutomatic in-flight checkpoint evaluation schedulingBenchmark evaluation frameworkCross-scaffold evaluationInline training evaluatorsJSON-structured answer-format promptmini-swe-agent harnessNeMo Gym and NeMo Evaluator-based harnessNeMo SkillsPool