docs: add MLflow integration documentation to fine-tuning examples

Add an MLflow Integration section to the READMEs for lora, osft, and
sft fine-tuning examples. Documents how to enable MLflow on the cluster
via the DataScienceCluster CR and MLflow CR, what the notebook cells do,
and how to navigate to the Experiments page in the RHOAI dashboard.

MLflow tracking is available for interactive (single-node) notebooks
only — distributed training does not currently support it.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: briangallagher

examples/fine-tuning/images/mlflow-experiments.png

@@ -172,3 +172,69 @@ to seamlessly run fine-tuning jobs.
 >   You can skip the token if switching to non-gated models.
 
 You can now proceed with the instructions from the notebook. Enjoy!
+
+## MLflow Integration (Optional)
+
+Training Hub supports [MLflow](https://mlflow.org/) for experiment tracking. When MLflow is enabled on your RHOAI cluster, training metrics (loss, learning rate, etc.) are automatically logged to MLflow experiments — no additional code changes required beyond setting the experiment name.
+
+> [!NOTE]
+> MLflow integration is available for **interactive (single node)** notebooks only. Distributed training jobs do not currently support MLflow tracking.
+
+### Enabling MLflow
+
+The interactive notebook already includes a cell that sets the MLflow experiment name:
+
+```python
+os.environ["MLFLOW_EXPERIMENT_NAME"] = "lora-training"
+```
+
+For this to work, MLflow must be enabled as a component in your RHOAI installation. If MLflow is not enabled, the environment variable is simply ignored and training proceeds normally.
+
+**To enable MLflow on your cluster:**
+
+1. Enable the MLflow Operator component in your `DataScienceCluster` CR:
+
+   ```bash
+   oc patch datasciencecluster default-dsc \
+     --type=merge \
+     -p '{"spec":{"components":{"mlflowoperator":{"managementState":"Managed"}}}}'
+   ```
+
+2. Create an `MLflow` CR to deploy the tracking server (example using SQLite and a PV for storage):
+
+   ```bash
+   oc apply -f - <<EOF
+   apiVersion: mlflow.opendatahub.io/v1
+   kind: MLflow
+   metadata:
+     name: mlflow
+   spec:
+     backendStoreUri: "sqlite:////mlflow/mlflow.db"
+     defaultArtifactRoot: "file:///mlflow/artifacts"
+     serveArtifacts: true
+     storage:
+       accessModes:
+         - ReadWriteOnce
+       resources:
+         requests:
+           storage: 10Gi
+   EOF
+   ```
+
+For full details, see the [Configuring MLflow in OpenShift AI](https://access.redhat.com/articles/7136121) Knowledgebase article.
+
+### Viewing MLflow Experiments
+
+Once training completes with MLflow enabled, you can browse your experiment runs:
+
+1. In the OpenShift AI dashboard, navigate to **Develop & train → Experiments** from the left sidebar menu.
+2. Select the experiment name (e.g., `lora-training`) to view all runs.
+3. Each run contains logged metrics (training loss, learning rate), parameters, and artifacts.
+
+You can also launch the full MLflow UI by clicking the **"Launch MLflow"** link in the top right of the Experiments page:
+
+![](../images/mlflow-experiments.png)
+
+Each run logs metrics including training loss, learning rate, samples per second, and more:
+
+![](../images/mlflow-run-metrics.png)

examples/fine-tuning/images/mlflow-run-metrics.png

@@ -207,3 +207,69 @@ These images serve both as training runtime and jupyter notebook images and come
 >   You can skip the token if switching to non-gated models.
 
 You can now proceed with the instructions from the notebook. Enjoy!
+
+## MLflow Integration (Optional)
+
+Training Hub supports [MLflow](https://mlflow.org/) for experiment tracking. When MLflow is enabled on your RHOAI cluster, training metrics (loss, learning rate, etc.) are automatically logged to MLflow experiments — no additional code changes required beyond setting the experiment name.
+
+> [!NOTE]
+> MLflow integration is available for **interactive (single node)** notebooks only. Distributed training jobs do not currently support MLflow tracking.
+
+### Enabling MLflow
+
+The interactive notebook already includes a cell that sets the MLflow experiment name:
+
+```python
+os.environ["MLFLOW_EXPERIMENT_NAME"] = "osft-training"
+```
+
+For this to work, MLflow must be enabled as a component in your RHOAI installation. If MLflow is not enabled, the environment variable is simply ignored and training proceeds normally.
+
+**To enable MLflow on your cluster:**
+
+1. Enable the MLflow Operator component in your `DataScienceCluster` CR:
+
+   ```bash
+   oc patch datasciencecluster default-dsc \
+     --type=merge \
+     -p '{"spec":{"components":{"mlflowoperator":{"managementState":"Managed"}}}}'
+   ```
+
+2. Create an `MLflow` CR to deploy the tracking server (example using SQLite and a PV for storage):
+
+   ```bash
+   oc apply -f - <<EOF
+   apiVersion: mlflow.opendatahub.io/v1
+   kind: MLflow
+   metadata:
+     name: mlflow
+   spec:
+     backendStoreUri: "sqlite:////mlflow/mlflow.db"
+     defaultArtifactRoot: "file:///mlflow/artifacts"
+     serveArtifacts: true
+     storage:
+       accessModes:
+         - ReadWriteOnce
+       resources:
+         requests:
+           storage: 10Gi
+   EOF
+   ```
+
+For full details, see the [Configuring MLflow in OpenShift AI](https://access.redhat.com/articles/7136121) Knowledgebase article.
+
+### Viewing MLflow Experiments
+
+Once training completes with MLflow enabled, you can browse your experiment runs:
+
+1. In the OpenShift AI dashboard, navigate to **Develop & train → Experiments** from the left sidebar menu.
+2. Select the experiment name (e.g., `osft-training`) to view all runs.
+3. Each run contains logged metrics (training loss, learning rate), parameters, and artifacts.
+
+You can also launch the full MLflow UI by clicking the **"Launch MLflow"** link in the top right of the Experiments page:
+
+![](../images/mlflow-experiments.png)
+
+Each run logs metrics including training loss, learning rate, samples per second, and more:
+
+![](../images/mlflow-run-metrics.png)

examples/fine-tuning/lora/README.md

@@ -154,3 +154,69 @@ to seamlessly run fine-tuning jobs.
 >   You can skip the token if switching to non-gated models.
 
 You can now proceed with the instructions from the notebook. Enjoy!
+
+## MLflow Integration (Optional)
+
+Training Hub supports [MLflow](https://mlflow.org/) for experiment tracking. When MLflow is enabled on your RHOAI cluster, training metrics (loss, learning rate, etc.) are automatically logged to MLflow experiments — no additional code changes required beyond setting the experiment name.
+
+> [!NOTE]
+> MLflow integration is available for **interactive (single node)** notebooks only. Distributed training jobs do not currently support MLflow tracking.
+
+### Enabling MLflow
+
+The interactive notebook already includes a cell that sets the MLflow experiment name:
+
+```python
+os.environ["MLFLOW_EXPERIMENT_NAME"] = "sft-training"
+```
+
+For this to work, MLflow must be enabled as a component in your RHOAI installation. If MLflow is not enabled, the environment variable is simply ignored and training proceeds normally.
+
+**To enable MLflow on your cluster:**
+
+1. Enable the MLflow Operator component in your `DataScienceCluster` CR:
+
+   ```bash
+   oc patch datasciencecluster default-dsc \
+     --type=merge \
+     -p '{"spec":{"components":{"mlflowoperator":{"managementState":"Managed"}}}}'
+   ```
+
+2. Create an `MLflow` CR to deploy the tracking server (example using SQLite and a PV for storage):
+
+   ```bash
+   oc apply -f - <<EOF
+   apiVersion: mlflow.opendatahub.io/v1
+   kind: MLflow
+   metadata:
+     name: mlflow
+   spec:
+     backendStoreUri: "sqlite:////mlflow/mlflow.db"
+     defaultArtifactRoot: "file:///mlflow/artifacts"
+     serveArtifacts: true
+     storage:
+       accessModes:
+         - ReadWriteOnce
+       resources:
+         requests:
+           storage: 10Gi
+   EOF
+   ```
+
+For full details, see the [Configuring MLflow in OpenShift AI](https://access.redhat.com/articles/7136121) Knowledgebase article.
+
+### Viewing MLflow Experiments
+
+Once training completes with MLflow enabled, you can browse your experiment runs:
+
+1. In the OpenShift AI dashboard, navigate to **Develop & train → Experiments** from the left sidebar menu.
+2. Select the experiment name (e.g., `sft-training`) to view all runs.
+3. Each run contains logged metrics (training loss, learning rate), parameters, and artifacts.
+
+You can also launch the full MLflow UI by clicking the **"Launch MLflow"** link in the top right of the Experiments page:
+
+![](../images/mlflow-experiments.png)
+
+Each run logs metrics including training loss, learning rate, samples per second, and more:
+
+![](../images/mlflow-run-metrics.png)