Open Models Almanac
Methodssoftware implementationtraining framework

single mention · software implementation

Experiments as Code

Experiments as Code expresses all inputs and configurations to any run as code committed into a single repository.1 Each run gets a unique ID and its inputs and artifacts are registered as assets with dependencies, giving reproducible history and end-to-end lineage.2,3

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. All inputs and configurations to any run (data pipelines, ablation experiments, any training run) are expressed as code committed into a single repository — laguna m1 xs2 technical report
  2. Each run is tracked with a unique ID, providing a stable handle for identifying, tracking, and reasoning about lineage across experiments and runs — laguna m1 xs2 technical report
  3. Keeping track of runs allows us to maintain a full reproducible history of what experiments were run, with what configuration and inputs, and where to find results and outputs — laguna m1 xs2 technical report

Filed under software implementation › training framework