Merge pull request #65 from nasa-nccs-hpda/ptc-refactor

Ptc refactor

.coveragerc

@@ -0,0 +1,25 @@
+[run]
+source = pytorch_caney 
+omit = 
+    */site-packages/*
+    */dist-packages/*
+    */tests/*
+    setup.py
+
+[report]
+exclude_lines =
+    pragma: no cover
+    def __repr__
+    if self.debug:
+    if __name__ == .__main__.:
+    raise NotImplementedError
+    pass
+    except ImportError:
+
+show_missing = True
+
+[html]
+directory = htmlcov
+
+[xml]
+output = coverage.xml

README.md

@@ -125,6 +125,18 @@ cd pytorch-caney; conda env create -f requirements/environment_gpu.yml;
 conda activate pytorch-caney
 python -m unittest discover pytorch_caney/tests
 ```
+
+Another example using the singularity exec command from the Explore system:
+
+```bash
+singularity exec --env PYTHONPATH="$NOBACKUP/development/pytorch-caney" --nv -B /path/to/mount /path/to/container/pytorch-caney-container coverage run -m unittest discover pytorch_caney/tests
+
+This command would output the report per file:
+
+```bash
+singularity exec --env PYTHONPATH="$NOBACKUP/development/pytorch-caney" --nv -B /path/to/mount /path/to/container/pytorch-caney-container coverage report
+```
+
 ## References
 
 - [Pytorch Lightning](https://github.com/Lightning-AI/lightning)

configs/3dcloudtask_fcn_baseline_test.yaml

@@ -0,0 +1,32 @@
+PIPELINE: '3dcloud'
+DATAMODULE: 'abitoa3dcloud'
+MODEL:
+  ENCODER: 'fcn'
+  DECODER: 'fcn'
+  NAME: 3dcloud-fcn-baseline
+  IN_CHANS: 14
+  DROP_PATH_RATE: 0.1
+DATA:
+  BATCH_SIZE: 32 
+  DATA_PATHS: [/explore/nobackup/projects/ilab/data/satvision-toa/3dcloud.data/abiChipsNew/]
+  TEST_DATA_PATHS: [/explore/nobackup/projects/ilab/data/satvision-toa/3dcloud.data/abiChipsNew/]
+  IMG_SIZE: 128 
+TRAIN:
+  ACCELERATOR: 'gpu'
+  STRATEGY: 'auto'
+  EPOCHS: 50 
+  WARMUP_EPOCHS: 10
+  BASE_LR: 3e-4 
+  MIN_LR: 2e-4
+  WARMUP_LR: 1e-4
+  WEIGHT_DECAY: 0.05
+  LR_SCHEDULER:
+    NAME: 'multistep'
+    GAMMA: 0.1
+    MULTISTEPS: [700,]
+LOSS:
+  NAME: 'bce'
+PRINT_FREQ: 10 
+SAVE_FREQ: 50
+VALIDATION_FREQ: 20
+TAG: 3dcloud_task_fcn_baseline_128_scaled_bt_minmax

configs/3dcloudtask_swinv2_satvision_giant_test.yaml

@@ -0,0 +1,41 @@
+PIPELINE: '3dcloud'
+DATAMODULE: 'abitoa3dcloud'
+MODEL:
+  ENCODER: 'satvision'
+  DECODER: 'fcn'
+  PRETRAINED: /panfs/ccds02/nobackup/projects/ilab/projects/3DClouds/models/SV-TOA/3B_2M/mp_rank_00_model_states.pt
+  TYPE: swinv2
+  NAME: 3dcloud-svtoa-finetune-giant
+  IN_CHANS: 14
+  DROP_PATH_RATE: 0.1
+  SWINV2:
+    IN_CHANS: 14
+    EMBED_DIM: 512
+    DEPTHS: [ 2, 2, 42, 2 ]
+    NUM_HEADS: [ 16, 32, 64, 128 ]
+    WINDOW_SIZE: 8
+    NORM_PERIOD: 6
+DATA:
+  BATCH_SIZE: 32 
+  DATA_PATHS: [/explore/nobackup/projects/ilab/data/satvision-toa/3dcloud.data/abiChipsNew/]
+  TEST_DATA_PATHS: [/explore/nobackup/projects/ilab/data/satvision-toa/3dcloud.data/abiChipsNew/]
+  IMG_SIZE: 128 
+TRAIN:
+  USE_CHECKPOINT: True
+  EPOCHS: 50 
+  WARMUP_EPOCHS: 10
+  BASE_LR: 3e-4 
+  MIN_LR: 2e-4
+  WARMUP_LR: 1e-4
+  WEIGHT_DECAY: 0.05
+  LR_SCHEDULER:
+    NAME: 'multistep'
+    GAMMA: 0.1
+    MULTISTEPS: [700,]
+LOSS:
+  NAME: 'bce'
+PRECISION: 'bf16'
+PRINT_FREQ: 10 
+SAVE_FREQ: 50
+VALIDATION_FREQ: 20
+TAG: 3dcloud_task_swinv2_g_satvision_128_scaled_bt_minmax

configs/mim_pretrain_swinv2_satvision_giant_128_onecycle_100ep.yaml

@@ -0,0 +1,48 @@
+PIPELINE: 'satvisiontoapretrain'
+
+MODEL:
+  TYPE: swinv2
+  NAME: mim_satvision_pretrain-giant
+  DROP_PATH_RATE: 0.1
+  SWINV2:
+    IN_CHANS: 14
+    EMBED_DIM: 512
+    DEPTHS: [ 2, 2, 42, 2 ]
+    NUM_HEADS: [ 16, 32, 64, 128 ]
+    WINDOW_SIZE: 8
+    NORM_PERIOD: 6
+
+DATA:
+  DATAMODULE: False
+  BATCH_SIZE: 64 
+  LENGTH: 1_920_000
+  PIN_MEMORY: True
+  NUM_WORKERS: 4
+  DATA_PATHS: [/explore/nobackup/projects/ilab/projects/3DClouds/data/mosaic-v3/webdatasets/shards]
+  IMG_SIZE: 128
+  MASK_PATCH_SIZE: 8
+  MASK_RATIO: 0.6
+
+TRAIN:
+  ACCELERATOR: 'gpu'
+  STRATEGY: 'deepspeed'
+  USE_CHECKPOINT: True
+  EPOCHS: 50 
+  WARMUP_EPOCHS: 10
+  BASE_LR: 3e-4
+  MIN_LR: 2e-4
+  WARMUP_LR: 1e-4
+  WEIGHT_DECAY: 0.05
+  LR_SCHEDULER:
+    NAME: 'multistep'
+    GAMMA: 0.1
+    MULTISTEPS: [700,]
+
+DEEPSPEED:
+  STAGE: 2
+
+PRECISION: 'bf16'
+
+PRINT_FREQ: 10 
+SAVE_FREQ: 50
+TAG: mim_pretrain_giant_satvision_128_scaled_bt_minmax_50ep

configs/mim_pretrain_swinv2_satvision_giant_128_onecycle_100ep_resume.yaml

@@ -0,0 +1,49 @@
+PIPELINE: 'satvisiontoapretrain'
+
+MODEL:
+  TYPE: swinv2
+  NAME: mim_satvision_pretrain-giant
+  DROP_PATH_RATE: 0.1
+  PRETRAINED: /panfs/ccds02/nobackup/projects/ilab/projects/3DClouds/models/SV-TOA/3B_2M/mp_rank_00_model_states.pt
+  SWINV2:
+    IN_CHANS: 14
+    EMBED_DIM: 512
+    DEPTHS: [ 2, 2, 42, 2 ]
+    NUM_HEADS: [ 16, 32, 64, 128 ]
+    WINDOW_SIZE: 8
+    NORM_PERIOD: 6
+
+DATA:
+  DATAMODULE: False
+  BATCH_SIZE: 64 
+  LENGTH: 1_920_000
+  PIN_MEMORY: True
+  NUM_WORKERS: 4
+  DATA_PATHS: [/explore/nobackup/projects/ilab/projects/3DClouds/data/mosaic-v3/webdatasets/shards]
+  IMG_SIZE: 128
+  MASK_PATCH_SIZE: 8
+  MASK_RATIO: 0.6
+
+TRAIN:
+  ACCELERATOR: 'gpu'
+  STRATEGY: 'deepspeed'
+  USE_CHECKPOINT: True
+  EPOCHS: 50 
+  WARMUP_EPOCHS: 10
+  BASE_LR: 3e-4
+  MIN_LR: 2e-4
+  WARMUP_LR: 1e-4
+  WEIGHT_DECAY: 0.05
+  LR_SCHEDULER:
+    NAME: 'multistep'
+    GAMMA: 0.1
+    MULTISTEPS: [700,]
+
+DEEPSPEED:
+  STAGE: 2
+
+PRECISION: 'bf16'
+
+PRINT_FREQ: 10 
+SAVE_FREQ: 50
+TAG: mim_pretrain_giant_satvision_128_scaled_bt_minmax_50ep_resume