Document serverless W&B run-finished alert behavior

docs/features/tracking-metrics.mdx

@@ -8,6 +8,13 @@ icon: "chart-line"
 ART writes a metrics row every time you call `model.log(...)`. Those rows go to
 `history.jsonl` in the run directory and, if W&B logging is enabled, to W&B.
 
+Serverless training also creates W&B-backed artifacts and runs for each remote
+training job so checkpoints can be traced back to their inputs. If W&B "Run
+finished" notifications are enabled for your account, a multi-step
+`ServerlessBackend` training loop can therefore send one notification per
+`backend.train(...)` call. See [ART Backend](/fundamentals/art-backend#serverlessbackend)
+for the serverless lifecycle notes and alert workaround.
+
 Use this page for three things:
 
 - understand the metrics ART emits automatically

docs/fundamentals/art-backend.mdx

@@ -55,6 +55,23 @@ backend = ServerlessBackend(
 
 As your training job progresses, `ServerlessBackend` automatically saves your LoRA checkpoints as W&B Artifacts and deploys them for production inference on W&B Inference.
 
+Each `backend.train(...)` call submits one remote training job. ART stores the
+job inputs and outputs in W&B so that every trained step has its own artifacts,
+metrics, and provenance. If your W&B user settings send Slack notifications
+when runs finish, that can produce one notification per training step in a loop
+such as:
+
+```python
+for step in range(num_steps):
+    groups = await art.gather_trajectory_groups(...)
+    result = await backend.train(model, groups, learning_rate=1e-5)
+```
+
+This is expected for the current serverless training lifecycle. To reduce alert
+noise, disable W&B "Run finished" notifications for the account or use a W&B
+account/team whose notification settings are dedicated to ART training jobs.
+ART still records checkpoints and provenance for each step.
+
 ### LocalBackend
 
 The `LocalBackend` class runs a vLLM server and either an Unsloth or torchtune instance on whatever machine your agent itself is executing. This is a good fit if you're already running your agent on a machine with a GPU.