Tutorials/AI-Workflows-Weights-and-Biases.md

-Weights & Biases is a tool for tracking machine learning experiments, visualizing and sharing results. 
-In the PIK cluster, where compute nodes have no internet access, syncing data to the W&B servers needs to be handled via the login node.
+Weights & Biases is a tool for tracking machine learning experiments, visualizing and sharing results. In the PIK cluster, where compute nodes have no internet access, syncing data to the W&B servers needs to be handled via the login node.
 
 ## Getting Started
+
 **Step 1: Create a W&B Account**
+
 - Academic researchers can apply for a free academic W&B account, which includes 100GB of free storage (https://wandb.ai/site/pricing/)
 
 **Step 2: Install W&B**
+
 - Install the W&B package using pip
 
 ```bash
@@ -13,6 +15,7 @@ pip install wandb
 ```
 
 **Step 3: Configure W&B**
+
 - Login to your W&B account using the following command
 
 ```bash
@@ -22,6 +25,7 @@ wandb login
 - Enter the API key, which can be obtained from: https://wandb.ai/authorize. You can also find it under user settings in your W&B account.
 
 **Step 4: Sync Data from the Login Node**
+
 - To synchronize data collected offline on compute nodes to the W&B servers, use the following command from the login node
 
 ```bash
@@ -30,7 +34,36 @@ wandb sync /path/to/offline-run-data
 
 ## PyTorch Integration Example
 
+**Step 1: Import W&B Library**
+
+```python
+import wandb
+```
+
+**Step 2: Initialize W&B**
+
+- Make sure to save data locally during training by setting the mode to `offline`
+
+```python
+wandb.init(project="my_project", entity="my_entity", mode="offline")
+```
+
+**Step 3: Training Loop**
+
+```python
+for epoch in range(num_epochs):
+    training_loss = 0
+    for data, target in enumerate(train_loader):
+        optimizer.zero_grad()
+        output = model(data)
+        loss = criterion(output, target)
+        loss.backward()
+        optimizer.step()
+        training_loss += loss.item()
+    # Log training metrics to W&B
+    wandb.log({"epoch": epoch, "train/loss": training_loss/len(trainloader)})
+```
+
 ## PyTorch Lightning Integration Example
 
 ## Uploading Model Checkpoints to W&B
\ No newline at end of file
-