adding configs to repo, and updating examples in readme

.gitignore

@@ -4,7 +4,7 @@
 # don't commit stuff in data, it may be too large and it's better to move it in other ways.
 # except the configs of your experiments
 data/*
-!data/configs/
+!data/config/
 
 # Don't commit rclone in case you are using it
 rclone.exe

README.md

@@ -90,6 +90,15 @@ python 01_train_model.py +config=alternative.yaml
 ###############################
 python 01_train_model.py --multirun model.num_layers=1,2,3
 
+###############################
+# Executing multiple runs with launchers
+###############################
+python 01_train_model.py --multirun model.num_layers=1,2,3 +launcher=joblib
+
+# or 
+
+python 01_train_model.py --multirun model.num_layers=1,2,3 +launcher=slurm
+
 ###############################
 # Using Hydra and Slurm for cluster job submissions
 ###############################

data/config/experiment/sweep_models_lr.yaml

+# @package _global_
+defaults:
+  - override /training: default
+
+hydra: 
+  mode: MULTIRUN
+  sweeper:
+    params:
+      model: net2,net5
+      training.lr: 1.0,0.1,0.001
+      training.epochs: 1

data/config/experiment/train_model_B.yaml

+# @package _global_
+defaults:
+  - override /model: net2
+  - override /training: default

data/config/launcher/joblib.yaml

+# @package _global_
+defaults:
+  - override /hydra/launcher: joblib
+
+hydra: 
+  launcher: # https://hydra.cc/docs/plugins/joblib_launcher/
+    n_jobs: 5
+

data/config/launcher/slurm.yaml

+# @package _global_
+defaults:
+  - override /hydra/launcher: submitit_slurm
+
+hydra: 
+  callbacks:
+    log_job_return:
+      _target_: hydra.experimental.callbacks.LogJobReturnCallback
+  launcher:
+    setup: [which python, echo 1]
+    submitit_folder: ${hydra.sweep.dir}/.submitit/%j
+      

data/config/launcher/submitit_local.yaml

+# @package _global_
+defaults:
+  - override /hydra/launcher: submitit_local 
+
+hydra: 
+  callbacks:
+    log_job_return:
+      _target_: hydra.experimental.callbacks.LogJobReturnCallback
+  launcher: # https://hydra.cc/docs/plugins/submitit_launcher/
+    submitit_folder: ${hydra.sweep.dir}/.submitit/%j
+    nodes: 1
+      

data/config/model/default.yaml

+# Model configuration for SimpleNet
+num_layers: 3  # Number of layers in the neural network (int)

data/config/model/net2.yaml

+# Model configuration for SimpleNet
+num_layers: 2  # Number of layers in the neural network (int)

data/config/model/net5.yaml

+# Model configuration for SimpleNet
+num_layers: 5  # Number of layers in the neural network (int)

data/config/model/net7.yaml

+# Model configuration for SimpleNet
+num_layers: 7  # Number of layers in the neural network (int)

data/config/train_model.yaml

+defaults:
+  - model: default
+  - training: default
+
+hydra:
+  run:
+    dir: ./data/outputs/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}
+  sweep:
+    dir: ./data/multirun/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}
+    subdir: ${hydra.job.num}
\ No newline at end of file

data/config/training/default.yaml

+# Training configuration for MNIST model
+batch_size: 512        # Input batch size for training (int)
+test_batch_size: 1000 # Input batch size for testing (int)
+epochs: 14            # Number of epochs to train (int)
+lr: 1.0               # Learning rate (float)
+gamma: 0.7            # Factor for the learning rate scheduler (float)
+no_cuda: false        # Disable CUDA (bool)
+no_mps: false         # Disable MacOS MPS GPU support (bool)
+dry_run: false        # Perform a dry run (do not update weights) (bool)
+seed: 1               # Seed for random number generation (int)
+log_interval: 10      # How often to log progress (int)
+save_model: true     # Whether to save the model to disk (bool)
+data_dir: './data/datasets'    # Directory to store the dataset (str)
+model_dir: './data/models'  # Directory to save trained models (str)