EvoGround: Agentic, evidence-grounded adaptation for rapidly shifting language. Detect when to adapt via multi-signal drift sensing, select what to learn, and update models with PEFT/LoRA—backed by prequential evaluation, fairness gates, and event-aware analysis.
EvoGround/
│
├── src/
│ ├── agentic/
│ ├── baseline/
│ ├── models/
│ ├── evaluation/
│ ├── data/
│ └── utils/
│
├── experiments/
│ ├── config/
│ ├── scripts/
│ ├── results/
│ └── notebooks/
│
├── cluster/ # HPC utilities
│ ├── trillium/
│ │ ├── job_train.slurm # SLURM script for Trillium
│ │ ├── job_eval.slurm
│ │ └── README.md # instructions + modules to load
│ ├── vector/
│ │ ├── job_train.slurm
│ │ ├── job_eval.slurm
│ │ └── README.md
│
├── envs/
│ ├── environment.yml # conda environment for training
│ ├── environment_min.yml # minimal environment
│ └── environment_eval.yml # inference/evaluation env
│
├── containers/ # For reproducibility
│ ├── Dockerfile
│ ├── evoground.sif # (not committed, ignored)
│ └── README.md
│
├── data/
│ ├── raw/ # large data -> ignored
│ ├── processed/
│ ├── monthly_splits/
│ ├── policy_events.csv
│ ├── covid_metrics.csv
│ └── README.md
│
├── reports/
│ ├── figs/
│ ├── tables/
│ └── experiment_summary.md
│
├── .gitignore
├── README.md
└── LICENSE
Distributed under the MIT License.
See LICENSE for details.