Skip to content

README.md

Latest commit

 

History

History
57 lines (37 loc) · 1.98 KB

README.md

File metadata and controls

57 lines (37 loc) · 1.98 KB

MultiGPrompt

  • Library code: pygfm/baseline_models/multigprompt/
  • Experiment driver: execute.py in this folder

Install

cd /path/to/repo
pip install -e .

Run from repository root:

python scripts/multigprompt/execute.py --dataset cora --gpu 0

Config files (YAML + Pydantic validation, recommended)

Long option lists can go in YAML and be passed with --config / -c; explicit CLI flags override the file.

pip install pydantic pyyaml

python scripts/multigprompt/execute.py --config configs/multigprompt/example_cora.yaml
python scripts/multigprompt/execute.py -c configs/multigprompt/example_cora.yaml --epochs 500 --gpu 1
  • Config schema: pygfm/public/cli/multigprompt_config.py (MultiGPromptRunConfig)
  • Example: configs/multigprompt/example_cora.yaml

Common flags: --epochs, --tasks, --inner_steps, --save_name (default ckpts/multigprompt/checkpoints/<dataset>.pkl).

Data layout (aligned with MDGPT: upstream / downstream split)

Kind Default path Notes
Upstream (Planetoid) datasets/multigprompt/<dataset>/ Per-dataset subdirs, e.g. .../cora/ind.cora.*
Downstream (few-shot + CSV) downstream_data/multigprompt/<dataset>/ e.g. .../cora/fewshot_cora/, cora_fewshot.csv; repo ships Cora only

See also:

  • datasets/multigprompt/README.md
  • downstream_data/multigprompt/README.md

Migrating from a single flat directory

If everything used to live in one folder:

  • Put ind.cora.* under datasets/multigprompt/cora/ (same idea for citeseer)
  • Put Cora few-shot assets under downstream_data/multigprompt/cora/
  • MULTIGPROMPT_*_DATA_DIR env vars point at a parent directory whose children are cora/, citeseer/, etc.—not a single dataset folder

Or keep MULTIGPROMPT_DATA_DIR pointing at the old combined directory for legacy behavior.

Optional extras: pip install -e ".[multigprompt]" (see root pyproject.toml).