Instructions to use msherry/pakflood-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use msherry/pakflood-lora with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Download code/h12_chain.sh from msherry/pakflood-lora: direct link, hf CLI and curl.
- Browser
- Download file 1.47 kB
-
https://huggingface.co/msherry/pakflood-lora/resolve/main/code/h12_chain.sh
- Command line
-
hf download hf://msherry/pakflood-lora/code/h12_chain.sh
-
curl -L -o h12_chain.sh https://huggingface.co/msherry/pakflood-lora/resolve/main/code/h12_chain.sh
1.47 kB
| # H1/H2 follow-up chain for lora_r32_all (rank search) and lora_r16_cedice (loss ablation). Queue ON THE SERVER: | |
| # bash ~/pakflood/code/pakflood/run_job.sh cmd h12_chain bash pakflood/h12_chain.sh | |
| # For each model: terratorch test on the official test split and on the PK official test (15 tiles), | |
| # full-tile predictions on the official test split (2417 tiles, for bootstrap CIs / precision-recall), | |
| # EMSR838 maps (177 tiles) + WorldPop population. ~15 min GPU. Log: ~/pakflood/jobs/h12_chain.log | |
| set -u | |
| cd "$HOME/pakflood/code" || exit 1 | |
| export HF_HOME="$HOME/pakflood/.cache/hf" CUDA_VISIBLE_DEVICES=1 PYTHONPATH="$HOME/pakflood/code" | |
| TT=.venv/bin/terratorch; PY=.venv/bin/python | |
| PREDS="$HOME/pakflood/preds" | |
| WP="$HOME/pakflood/data/worldpop/pak_ppp_2020_constrained.tif" | |
| run() { echo; echo "#### $* ($(date +%H:%M))"; "$@"; echo "#### rc=$?"; } | |
| for m in lora_r32_all lora_r16_cedice; do | |
| CFG=pakflood/configs/$m.yaml | |
| CK=workdir/pakflood/$m/v1/checkpoints/best.ckpt | |
| run $TT test --config $CFG --ckpt_path $CK | |
| run $TT test --config $CFG --ckpt_path $CK --data.init_args.metadata_filename data_tiles_pk_test.csv | |
| run $PY pakflood/predict_tiles.py $CK $PREDS/${m}_test --events all --split test --task pakflood | |
| run $PY pakflood/predict_tiles.py $CK $PREDS/${m}_emsr838 --events EMSR838 --split all --task pakflood | |
| run $PY pakflood/population.py $PREDS/${m}_emsr838 "$WP" --out $PREDS/pop_$m.csv | |
| done | |
| echo "#### chain done $(date)" | |