Heuristic:Allenai Open instruct Warning Archived Dev Scripts
| Knowledge Sources | |
|---|---|
| Domains | Training, Hyperparameter_Search |
| Last Updated | 2026-02-07 03:00 GMT |
Overview
Deprecation warning: scripts in docs/archived_dev_scripts/ are historical records not intended for direct execution.
Description
The archived_dev_scripts directory contains shell scripts that document past hyperparameter sweeps and training pipelines for specific model families (OLMo 2, OLMoE). These scripts reference hardcoded checkpoint paths, specific Beaker cluster configurations, and WandB project names that may no longer be valid. They are preserved as reference material for reproducing published results or designing new experiments.
Usage
Consult this heuristic when encountering scripts from docs/archived_dev_scripts/. Do not execute these scripts directly. Instead, use them as templates: extract individual commands, update model paths, cluster names, and dataset references to match current infrastructure before running.
The Insight (Rule of Thumb)
- Action: Do not run archived dev scripts directly. Copy individual commands and update paths/configs.
- Value: Paths like /weka/... and cluster names like ai2/jupiter-cirrascale-2 may be stale.
- Trade-off: Using archived scripts as-is risks failed jobs due to missing models, datasets, or cluster resources.
Reasoning
These scripts were checked into docs/archived_dev_scripts/ explicitly to serve as historical records. The directory name signals their archived status. Infrastructure changes (cluster migrations, model relocations, dataset updates) mean concrete values in these scripts drift over time. The scripts remain valuable as documentation of experimental methodology and hyperparameter ranges.