[2.6] Update WandB code and example (#3429)

Use job api for WandB example.
Fix WandBReceiver bug.

### Description

Use job api for WandB example.
Fix WandBReceiver bug.

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [x] Non-breaking change (fix or new feature that would not break
existing functionality).
- [ ] Breaking change (fix or new feature that would cause existing
functionality to change).
- [ ] New tests added to cover the changes.
- [ ] Quick tests passed locally by running `./runtest.sh`.
- [ ] In-line docstrings updated.
- [ ] Documentation updated.

---------

Co-authored-by: Chester Chen <[email protected]>

Author: YuanTingHsieh

examples/advanced/experiment-tracking/wandb/README.md

@@ -8,19 +8,14 @@ This example also highlights the Weights and Biases streaming capability from th
 
 > **_NOTE:_** This example uses the [CIFAR-10](https://www.cs.toronto.edu/~kriz/cifar.html) dataset and will load its data within the trainer code.
 
-### 1. Install requirements and configure PYTHONPATH
+### 1. Install requirements
 
 Install additional requirements (if you already have a specific version of nvflare installed in your environment, you may want to remove nvflare in the requirements to avoid reinstalling nvflare):
 
 ```
 python -m pip install -r requirements.txt
 ```
 
-Set `PYTHONPATH` to include custom files of this example:
-```
-export PYTHONPATH=${PWD}/..
-```
-
 ### 2. Make sure the FL server is logged into Weights and Biases
 
 Import the W&B Python SDK and log in:
@@ -35,42 +30,51 @@ Provide your API key when prompted.
 
 ### 3. Run the experiment
 
-Use nvflare simulator to run the example:
+Use job api to run the example:
 
 ```
-nvflare simulator -w /tmp/nvflare/ -n 2 -t 2 ./jobs/hello-pt-wandb
+python wandb_job.py
 ```
 
-### 3. Access the logs and results
+### 4. Access the logs and results
 
 By default, Weights and Biases will create a directory named "wandb" in the server workspace. With "mode": "online" in the WandBReceiver, the
 files will be synced with the Weights and Biases server. You can visit https://wandb.ai/ and log in to see your run data.
 
-### 4. Weights and Biases tracking
+### 5. How it works
 
-For the job `hello-pt-wandb`, on the client side, the client code in `PTLearner` uses the syntax for Weights and Biases:
+To enable tracking with Weights & Biases (WandB), you can use the `WandBWriter` utility provided by NVFlare. Here's a basic example of how to integrate it into your training script:
 
-```
-self.writer.log({"train_loss": cost.item()}, current_step)
+```python
+from nvflare.client.tracking import WandBWriter
+
+wandb_writer = WandBWriter()
+wandb_writer.log({"train_loss": cost.item()}, current_step)
 
-self.writer.log({"validation_accuracy": metric}, epoch)
 ```
 
-The `WandBWriter` mimics the syntax from Weights and Biases to send the information in events to the server through NVFlare events
-of type `analytix_log_stats` for the server to write the data for the WandB tracking server.
+The `WandBWriter` follows a syntax similar to the native WandB API, making it easy to adopt.
+
+Internally, `WandBWriter` leverages the NVFlare client API to send metrics and trigger an `analytix_log_stats` event. This event can be received and processed by our `AnalyticsReceiver`, with the `WandBReceiver` being one implementation of it.
 
-The `ConvertToFedEvent` widget turns the event `analytix_log_stats` into a fed event `fed.analytix_log_stats`,
-which will be delivered to the server side.
+In `wandb_job.py`, we configure the following components by default:
 
-On the server side, the `WandBReceiver` is configured to process `fed.analytix_log_stats` events,
-which writes received data from these events.
+  - The `ConvertToFedEvent` widget on the NVFlare client side, which transfroms the event `analytix_log_stats` into a fed event `fed.analytix_log_stats`. This enables the event to be sent from the NVFlare client to the NVFlare server.
 
-This allows for the server to be the only party that needs to deal with authentication for the WandB tracking server, and the server
-can buffer the events from many clients to better manage the load of requests to the tracking server.
+  - The `WandBReceiver` on the NVFlare server side, which listens for `fed.analytix_log_stats` events and forwards the metric data to the WandB tracking server.
 
-### 5. Sends to WandB server directly from client side
+This setup ensures that the server handles all authentication with the WandB tracking server and buffers events from multiple clients, effectively managing the load of requests to the server.
+
+### 6. Optional: Stream Metrics Directly from Clients
+
+Alternatively, you can stream metrics **directly from each NVFlare client** to WandB, bypassing the NVFlare server entirely.
+
+To enable this mode, run your training script with the following flags:
+
+```bash
+python wandb_job.py --streamed_to_clients --no-streamed_to_server
+```
 
-You can stream the metrics to the MLFlow server without passing through the NVFlare server as well.
-Please check the job `hello-pt-wandb-client`.
+In this configuration, the `WandBReceiver` is set up on the NVFlare client side to process the `analytix_log_stats` event.
 
-You notice we configure the `WandBReceiver` on the client side to process the `analytix_log_stats` event.
+As a result, each NVFlare client sends its metrics directly to its corresponding WandB server.