Recipe catalog#

pimm ships 22 recipes. A recipe is a config under configs/; pass its name to --train.config. uv run pimm ls prints the same names. Batch size counts events per step across all GPUs. A blank warm start means the run starts from random weights unless you pass --train.weight.

detector/semseg#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

detector/semseg/semseg-pt-v3m2-jaxtpc-5cls

PTv3 semantic segmentation on JAXTPC 3D data.

DefaultSegmentorV2 · PT-v3m2

JAXTPCDataset

—

20

48

panda/panseg#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

panda/panseg/detector-v5-pt-v3m2-ft-joint-fft

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v3

—

20

48

panda/panseg/detector-v5-pt-v3m2-ft-pid-dec

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v2

—

20

48

panda/panseg/detector-v5-pt-v3m2-ft-pid-fft-detector

Full fine-tuning from the published Panda Particle detector.

detector-v5 · PT-v3m2

PILArNetH5Dataset v2

hf://DeepLearnPhysics/panda-particle@bd90792dfe83cd05b437b719564b311f0a0b785a

20

48

panda/panseg/detector-v5-pt-v3m2-ft-pid-fft

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v2

—

20

48

panda/panseg/detector-v5-pt-v3m2-ft-pid-scratch

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v2

—

20

48

panda/panseg/detector-v5-pt-v3m2-ft-vtx-dec

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v3

—

20

48

panda/panseg/detector-v5-pt-v3m2-ft-vtx-fft-detector

Full fine-tuning from the published Panda Interaction detector.

detector-v5 · PT-v3m2

PILArNetH5Dataset v3

hf://DeepLearnPhysics/panda-interaction@fa8bd4c1937c39a48e02bde5bf27e188d0be4ac4

20

48

panda/panseg/detector-v5-pt-v3m2-ft-vtx-fft

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v3

—

20

48

panda/panseg/detector-v5-pt-v3m2-ft-vtx-scratch

—

detector-v5 · PT-v3m2

PILArNetH5Dataset v3

—

20

48

panda/pretrain#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

panda/pretrain/pretrain-sonata-v1m1-pilarnet-smallmask-v3m8

Configuration for pretraining a SONATA model on PILArNet dataset with PT-v3m8 backbone (bottleneck CPE variant of Utonia)

Sonata-v1m1 · PT-v3m8

PILArNetH5Dataset v2

—

100

48

panda/pretrain/pretrain-sonata-v1m1-pilarnet-smallmask

Configuration for pretraining a SONATA model on PILArNet dataset

Sonata-v1m1 · PT-v3m2

PILArNetH5Dataset v1

—

100

48

panda/semseg#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

panda/semseg/semseg-pt-v3m2-pilarnet-ft-5cls-dec

—

DefaultSegmentorV2 · PT-v3m2

PILArNetH5Dataset v1

—

20

48

panda/semseg/semseg-pt-v3m2-pilarnet-ft-5cls-fft

—

DefaultSegmentorV2 · PT-v3m2

PILArNetH5Dataset v1

—

20

48

panda/semseg/semseg-pt-v3m2-pilarnet-ft-5cls-lin

—

DefaultSegmentorV2 · PT-v3m2

PILArNetH5Dataset v1

—

20

48

panda/semseg/semseg-pt-v3m2-pilarnet-ft-5cls-scratch

—

DefaultSegmentorV2 · PT-v3m2

PILArNetH5Dataset v1

—

20

48

polarmae#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

polarmae/pretrain-polarmae-pilarnet

Self-contained PoLAr-MAE pre-training on PILArNet.

PoLAr-MAE

PILArNetH5Dataset

—

100

64

polarmae/semseg#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

polarmae/semseg/semseg-polarmae-pilarnet-fft-reproduce

PoLAr-MAE Semantic Segmentation - Reproduce fine-tuned checkpoint results.

PoLArMAE-SemSeg

PILArNetH5Dataset v1

—

1

32

polarmae/semseg/semseg-polarmae-pilarnet-fft

PoLAr-MAE Semantic Segmentation - Full Fine-Tuning (FFT)

PoLArMAE-SemSeg

PILArNetH5Dataset v1

—

100

32

polarmae/semseg/semseg-polarmae-pilarnet-peft

PoLAr-MAE Semantic Segmentation - Parameter-Efficient Fine-Tuning (PEFT)

PoLArMAE-SemSeg

PILArNetH5Dataset v1

—

50

48

tests#

Recipe

About

Model

Data

Warm start

Epochs

Batch size

tests/tiny_semseg

—

DefaultSegmentorV2 · PT-v3m2

PILArNetH5Dataset v2

—

1

4

tests/tiny_semseg_crash

—

DefaultSegmentorV2 · PT-v3m2

PILArNetH5Dataset v2

—

2

4

Recipes that don’t load#

These configs fail in Config.fromfile, so neither the catalog nor the launcher can use them.

Recipe

Error

panda/pretrain/pretrain-sonata-v1m3-pilarnet-smallmask-litept-v1m2-M-observable-parquet-stream

FileNotFoundError: file “configs/panda/pretrain/pretrain-sonata-v1m3-pilarnet-smallmask-litept-v1m2-M-observable.py” does not exist

panda/pretrain/pretrain-sonata-v1m3-pilarnet-smallmask-litept-v1m2-M-observable-parquet

FileNotFoundError: file “configs/panda/pretrain/pretrain-sonata-v1m3-pilarnet-smallmask-litept-v1m2-M-observable.py” does not exist