test: add unit tests and docs for Trackio monitor

- Add TestTrackio covering config defaults, enabled config, and
  write_events() logging behavior (mocks the optional trackio dependency)
- Add TestMonitorMasterTrackioWiring covering MonitorMaster registration
  when trackio is enabled/disabled
- Update docs/_tutorials/monitor.md with Trackio overview, config
  example, and Custom Monitoring API reference

Addresses Codex review feedback on PR #8065

Signed-off-by: Pranav Chanduri <preethivardhanc@gmail.com>

Author: chanduripranav

docs/_tutorials/monitor.md

@@ -11,7 +11,7 @@ In this tutorial, we introduce the DeepSpeed Monitor and provide examples of its
 
 ## Overview
 
-Monitoring model and system metrics during training is vital to ensure hardware resources are fully utilized. The DeepSpeed Monitor enables live logging of metrics through one or more monitoring backends such as PyTorch's [TensorBoard](https://pytorch.org/docs/1.8.0/tensorboard.html), [WandB](https://docs.wandb.ai/quickstart), [Comet](https://www.comet.com/site/?utm_source=deepseed&utm_medium=docs&utm_content=tutorial) and simple CSV files.
+Monitoring model and system metrics during training is vital to ensure hardware resources are fully utilized. The DeepSpeed Monitor enables live logging of metrics through one or more monitoring backends such as PyTorch's [TensorBoard](https://pytorch.org/docs/1.8.0/tensorboard.html), [WandB](https://docs.wandb.ai/quickstart), [Comet](https://www.comet.com/site/?utm_source=deepseed&utm_medium=docs&utm_content=tutorial), [Trackio](https://github.com/huggingface/trackio) and simple CSV files.
 
 Below is a live monitoring view for TensorBoard:
 
@@ -25,6 +25,8 @@ Below is a live monitoring view for Comet:
 
 ![CometML Example Output](/assets/images/comet_monitor.png){: .align-center}
 
+[Trackio](https://github.com/huggingface/trackio) is a lightweight, offline-first experiment tracking library from Hugging Face with a WandB-compatible API. Runs can be visualized as an HF Space or dataset on the HF Hub.
+
 ## Usage
 
 The DeepSpeed Monitor is configured within the deepspeed [configuration file](/docs/config-json/#monitoring-module). DeepSpeed will automatically monitor key training metrics, including those tracked with the `wall_clock_breakdown` configuration option. In addition, users can log their own custom events and metrics.
@@ -54,6 +56,10 @@ When using DeepSpeed for model training, the Monitor can be configured in the De
     "project": "my_project",
     "experiment_name": "my_experiment"
   }
+  "trackio": {
+    "enabled": true,
+    "project": "my_project"
+  }
   "csv_monitor": {
     "enabled": true,
     "output_path": "output/ds_logs/",
@@ -69,7 +75,7 @@ DeepSpeed will automatically log to all available and enabled monitoring backend
 In addition to automatic monitoring, users can log their own custom metrics in client scripts. Currently, there are two ways to initialize Monitor objects:
 
 1. (Recommended) - Create a `MonitorMaster(ds_config.monitor_config)` object, which automatically initializes all monitor backends present in the DeepSpeed configuration
-2. Create a specific `TensorBoardMonitor(ds_config.monitor_config)`, `WandbMonitor(ds_config.monitor_config)`, `csvMonitor(ds_config.monitor_config)` object which will only initialize a specific monitor backend present in the DeepSpeed configuration
+2. Create a specific `TensorBoardMonitor(ds_config.monitor_config)`, `WandbMonitor(ds_config.monitor_config)`, `TrackioMonitor(ds_config.monitor_config)`, `csvMonitor(ds_config.monitor_config)` object which will only initialize a specific monitor backend present in the DeepSpeed configuration
 
 
 The steps to create a custom monitor are as follows:

tests/unit/monitor/test_monitor.py

@@ -3,14 +3,18 @@
 
 # DeepSpeed Team
 
+import sys
+
 from deepspeed.monitor.tensorboard import TensorBoardMonitor
 from deepspeed.monitor.wandb import WandbMonitor
 from deepspeed.monitor.csv_monitor import csvMonitor
 from deepspeed.monitor.config import DeepSpeedMonitorConfig
 from deepspeed.monitor.comet import CometMonitor
+from deepspeed.monitor.trackio import TrackioMonitor
+from deepspeed.monitor.monitor import MonitorMaster
 
 from unit.common import DistributedTest
-from unittest.mock import Mock, patch
+from unittest.mock import Mock, MagicMock, patch
 from deepspeed.runtime.config import DeepSpeedConfig
 
 import deepspeed.comm as dist
@@ -164,3 +168,83 @@ def test_empty_comet(self):
         assert comet_monitor.enabled == defaults.enabled
         assert comet_monitor.samples_log_interval == defaults.samples_log_interval
         mock_start.assert_not_called()
+
+
+class TestTrackio(DistributedTest):
+    world_size = 2
+
+    def test_trackio(self):
+        # trackio is an optional dependency, so we stub the module rather
+        # than requiring it to be installed for CI.
+        mock_trackio = MagicMock()
+
+        config_dict = {"train_batch_size": 2, "trackio": {"enabled": True, "project": "my_project"}}
+        ds_config = DeepSpeedConfig(config_dict)
+
+        with patch.dict(sys.modules, {"trackio": mock_trackio}):
+            trackio_monitor = TrackioMonitor(ds_config.monitor_config.trackio)
+
+        assert trackio_monitor.enabled == True
+        assert trackio_monitor.project == "my_project"
+
+        # trackio.init should only be called on rank 0
+        if dist.get_rank() == 0:
+            mock_trackio.init.assert_called_once_with(project="my_project")
+        else:
+            mock_trackio.init.assert_not_called()
+
+    def test_empty_trackio(self):
+        mock_trackio = MagicMock()
+
+        config_dict = {"train_batch_size": 2, "trackio": {}}
+        ds_config = DeepSpeedConfig(config_dict)
+
+        with patch.dict(sys.modules, {"trackio": mock_trackio}):
+            trackio_monitor = TrackioMonitor(ds_config.monitor_config.trackio)
+
+        defaults = DeepSpeedMonitorConfig().trackio
+        assert trackio_monitor.enabled == defaults.enabled
+        assert trackio_monitor.project == defaults.project
+
+    def test_trackio_write_events(self):
+        # Verifies write_events() correctly converts 3-tuples into
+        # trackio.log() calls with the right step value.
+        mock_trackio = MagicMock()
+
+        config_dict = {"train_batch_size": 2, "trackio": {"enabled": True, "project": "my_project"}}
+        ds_config = DeepSpeedConfig(config_dict)
+
+        with patch.dict(sys.modules, {"trackio": mock_trackio}):
+            trackio_monitor = TrackioMonitor(ds_config.monitor_config.trackio)
+            events = [("Train/Loss", 0.5, 100)]
+            trackio_monitor.write_events(events)
+
+        if dist.get_rank() == 0:
+            mock_trackio.log.assert_called_once_with({"Train/Loss": 0.5}, step=100)
+        else:
+            mock_trackio.log.assert_not_called()
+
+
+class TestMonitorMasterTrackioWiring(DistributedTest):
+    world_size = 2
+
+    def test_trackio_enabled_creates_monitor(self):
+        mock_trackio = MagicMock()
+
+        config_dict = {"train_batch_size": 2, "trackio": {"enabled": True, "project": "my_project"}}
+        ds_config = DeepSpeedConfig(config_dict)
+
+        with patch.dict(sys.modules, {"trackio": mock_trackio}):
+            monitor_master = MonitorMaster(ds_config.monitor_config)
+
+        if dist.get_rank() == 0:
+            assert monitor_master.trackio_monitor is not None
+            assert isinstance(monitor_master.trackio_monitor, TrackioMonitor)
+        else:
+            assert monitor_master.trackio_monitor is None
+
+    def test_trackio_disabled_skips_monitor(self):
+        config_dict = {"train_batch_size": 2, "trackio": {"enabled": False}}
+        ds_config = DeepSpeedConfig(config_dict)
+        monitor_master = MonitorMaster(ds_config.monitor_config)
+        assert monitor_master.trackio_monitor is None