diff --git a/docs/source/guides/execution.md b/docs/source/guides/execution.md
index 1eb8d82e..d9945626 100644
--- a/docs/source/guides/execution.md
+++ b/docs/source/guides/execution.md
@@ -205,12 +205,23 @@ def your_skypilot_executor(nodes: int, devices: int, container_image: str):
     return SkypilotExecutor(
         gpus="RTX5880-ADA-GENERATION",
         gpus_per_node=devices,
-        nodes = nodes
-        env_vars=common_envs()
+        num_nodes = nodes,
+        env_vars=common_envs(),
         container_image=container_image,
-        cloud="kubernetes",
+        infra="k8s/mycontext",
         # Optional to reuse Skypilot cluster
         cluster_name="tester",
+        volumes={
+            "nemo-workspace": "nemo-workspace"
+        },
+        volume_mounts=[
+            {
+                "path": "/data",
+                "volume_name": "nemo-workspace",
+                "size": "50Gi",
+                "type": "k8s-pvc"
+            }
+        ],
         setup="""
     conda deactivate
     nvidia-smi
diff --git a/nemo_run/core/execution/skypilot.py b/nemo_run/core/execution/skypilot.py
index d3823c43..a60fa2b4 100644
--- a/nemo_run/core/execution/skypilot.py
+++ b/nemo_run/core/execution/skypilot.py
@@ -18,7 +18,7 @@
 import subprocess
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Any, Optional, Type, Union
+from typing import Any, Dict, List, Optional, Type, Union
 
 from invoke.context import Context
 
@@ -36,6 +36,8 @@
     import sky.task as skyt
     from sky import backends
     from sky.utils import status_lib
+    from sky.volumes import volume as volume_lib
+    from sky import models
 
     _SKYPILOT_AVAILABLE = True
 except ImportError:
@@ -94,6 +96,8 @@ class SkypilotExecutor(Executor):
     memory: Optional[Union[int | float, list[int | float]]] = None
     instance_type: Optional[Union[str, list[str]]] = None
     num_nodes: int = 1
+    volumes: Optional[Dict[str, str]] = None
+    volume_mounts: Optional[List[Any]] = None
     use_spot: Optional[Union[bool, list[bool]]] = None
     disk_size: Optional[Union[int, list[int]]] = None
     disk_tier: Optional[Union[str, list[str]]] = None
@@ -341,6 +345,64 @@ def macro_values(self) -> Optional[ExecutorMacros]:
             het_group_host_var=self.HET_GROUP_HOST_VAR,
         )
 
+    def _setup_launcher(self):
+        # Auto-enable torchrun for distributed training scenarios:
+        # 1. Multi-node training (num_nodes > 1)
+        # 2. Single-node multi-GPU training (gpus_per_node > 1)
+        if self.launcher is None and (
+            self.num_nodes > 1 or (self.gpus_per_node and self.gpus_per_node > 1)
+        ):
+            self.launcher = "torchrun"
+
+        super()._setup_launcher()
+
+    def supports_launcher_transform(self) -> bool:
+        return True
+
+    def _parse_infra_for_volume_config(self) -> dict:
+        """Parse infra string and return volume config parameters."""
+        config = {}
+
+        if self.infra is not None:
+            # Parse infra string to extract cloud, region, zone components
+            # Format: cloud, cloud/region, cloud/region/zone, k8s/context
+            infra_parts = self.infra.split("/")
+            cloud = infra_parts[0] if infra_parts else None
+
+            if cloud:
+                # Special handling for Kubernetes
+                if cloud == "k8s":
+                    # VolumeConfig region and zone required even though they are marked as optional
+                    # validation fails otherwise
+                    config["cloud"] = "kubernetes"
+                    config["region"] = "kubernetes"
+                    config["zone"] = "kubernetes"
+                else:
+                    # Handle regular cloud providers
+                    config["cloud"] = cloud
+
+                    # Handle region for non-k8s clouds
+                    if len(infra_parts) >= 2:
+                        region = infra_parts[1]
+                        if region and region != "*":  # Skip wildcards
+                            config["region"] = region
+
+                    # Handle zone for non-k8s clouds
+                    if len(infra_parts) >= 3:
+                        zone = infra_parts[2]
+                        if zone and zone != "*":  # Skip wildcards
+                            config["zone"] = zone
+        else:
+            # Fall back to individual cloud, region, zone parameters
+            if self.cloud:
+                config["cloud"] = self.cloud
+            if self.region:
+                config["region"] = self.region
+            if self.zone:
+                config["zone"] = self.zone
+
+        return config
+
     def to_task(
         self,
         name: str,
@@ -364,16 +426,43 @@ def to_task(
 
 {" ".join(cmd)}
 """
+
         task = Task(
             name=name,
             setup=self.setup if self.setup else "",
             run=run_cmd,
             envs=self.env_vars,
             num_nodes=self.num_nodes,
+            volumes=self.volumes,
         )
+
         file_mounts = self.file_mounts or {}
         file_mounts["/nemo_run"] = self.job_dir
         task.set_file_mounts(file_mounts)
+        task.set_volumes(self.volumes)
+
+        volume_mounts = []
+        if self.volume_mounts:
+            for volume_mount in self.volume_mounts:
+                # Configure volume based on infra if specified, otherwise use cloud/region/zone
+                volume_config_kwargs = {
+                    "name": volume_mount["volume_name"],
+                    "type": volume_mount["type"],
+                    "name_on_cloud": volume_mount["volume_name"],
+                    "size": volume_mount["size"],
+                }
+
+                # Add parsed infra configuration
+                volume_config_kwargs.update(self._parse_infra_for_volume_config())
+
+                volume_mounts.append(
+                    volume_lib.VolumeMount(
+                        path=volume_mount["path"],
+                        volume_name=volume_mount["volume_name"],
+                        volume_config=models.VolumeConfig(**volume_config_kwargs),
+                    )
+                )
+        task.volume_mounts = volume_mounts
         task.set_resources(self.to_resources())
 
         if env_vars:
diff --git a/test/core/execution/test_skypilot.py b/test/core/execution/test_skypilot.py
index fe975049..ee3f1e40 100644
--- a/test/core/execution/test_skypilot.py
+++ b/test/core/execution/test_skypilot.py
@@ -561,3 +561,233 @@ def test_to_task(self, mock_task, mock_skypilot_imports, executor):
 
                 # Verify the returned task is our mock
                 assert result == mock_task_instance
+
+    def test_parse_infra_for_volume_config(self, mock_skypilot_imports):
+        """Test the _parse_infra_for_volume_config helper method."""
+
+        # Test k8s infra
+        executor_k8s = SkypilotExecutor(infra="k8s/my-context")
+        config = executor_k8s._parse_infra_for_volume_config()
+        assert config["cloud"] == "kubernetes"
+        assert config["region"] == "kubernetes"
+        assert config["zone"] == "kubernetes"
+
+        # Test AWS infra with region and zone
+        executor_aws = SkypilotExecutor(infra="aws/us-east-1/us-east-1a")
+        config = executor_aws._parse_infra_for_volume_config()
+        assert config["cloud"] == "aws"
+        assert config["region"] == "us-east-1"
+        assert config["zone"] == "us-east-1a"
+
+        # Test fallback to individual parameters
+        executor_fallback = SkypilotExecutor(
+            cloud="gcp", region="us-central1", zone="us-central1-a"
+        )
+        config = executor_fallback._parse_infra_for_volume_config()
+        assert config["cloud"] == "gcp"
+        assert config["region"] == "us-central1"
+        assert config["zone"] == "us-central1-a"
+
+    def test_volume_mounts_initialization(self, mock_skypilot_imports):
+        """Test that volume_mounts are properly stored during initialization."""
+        volume_mounts = [
+            {"path": "/data", "volume_name": "nemo-workspace", "size": "50Gi", "type": "k8s-pvc"}
+        ]
+
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            cloud="kubernetes",
+            cluster_name="test-cluster",
+            volume_mounts=volume_mounts,
+        )
+
+        # Verify volume_mounts are stored correctly
+        assert executor.volume_mounts == volume_mounts
+        assert len(executor.volume_mounts) == 1
+        assert executor.volume_mounts[0]["path"] == "/data"
+        assert executor.volume_mounts[0]["volume_name"] == "nemo-workspace"
+
+    def test_volume_mounts_none(self, mock_skypilot_imports):
+        """Test that volume_mounts can be None."""
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            cloud="kubernetes",
+            cluster_name="test-cluster",
+            volume_mounts=None,
+        )
+
+        # Verify volume_mounts is None
+        assert executor.volume_mounts is None
+
+    @patch("sky.task.Task")
+    @patch("sky.volumes.volume.VolumeMount")
+    @patch("sky.models.VolumeConfig")
+    def test_volume_mounts_to_task_processing(
+        self, mock_volume_config, mock_volume_mount, mock_task, mock_skypilot_imports
+    ):
+        """Test that volume_mounts are processed in to_task method."""
+        mock_task_instance = MagicMock()
+        mock_task.return_value = mock_task_instance
+
+        volume_mounts = [
+            {"path": "/data", "volume_name": "nemo-workspace", "size": "50Gi", "type": "k8s-pvc"}
+        ]
+
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            cloud="kubernetes",
+            region="kubernetes",
+            zone="kubernetes",
+            cluster_name="test-cluster",
+            volume_mounts=volume_mounts,
+        )
+
+        with patch.object(SkypilotExecutor, "to_resources") as mock_to_resources:
+            mock_to_resources.return_value = MagicMock()
+
+            with tempfile.TemporaryDirectory() as tmp_dir:
+                executor.job_dir = tmp_dir
+                executor.to_task("test_task", ["python", "script.py"])
+
+                # Verify volume processing was called (covers the processing logic)
+                mock_volume_mount.assert_called_once()
+                mock_volume_config.assert_called_once()
+
+    @patch("sky.task.Task")
+    @patch("sky.volumes.volume.VolumeMount")
+    @patch("sky.models.VolumeConfig")
+    def test_volume_mounts_with_infra(
+        self, mock_volume_config, mock_volume_mount, mock_task, mock_skypilot_imports
+    ):
+        """Test volume_mounts processing when using infra instead of cloud/region/zone."""
+        mock_task_instance = MagicMock()
+        mock_task.return_value = mock_task_instance
+
+        volume_mounts = [
+            {"path": "/data", "volume_name": "nemo-workspace", "size": "50Gi", "type": "k8s-pvc"}
+        ]
+
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            infra="k8s/kubernetes-admin@kubernetes",
+            cluster_name="test-cluster",
+            volume_mounts=volume_mounts,
+        )
+
+        with patch.object(SkypilotExecutor, "to_resources") as mock_to_resources:
+            mock_to_resources.return_value = MagicMock()
+
+            with tempfile.TemporaryDirectory() as tmp_dir:
+                executor.job_dir = tmp_dir
+                executor.to_task("test_task", ["python", "script.py"])
+
+                # Verify volume processing was called (covers the infra parsing logic)
+                mock_volume_mount.assert_called_once()
+                mock_volume_config.assert_called_once()
+
+    @patch("sky.task.Task")
+    @patch("sky.volumes.volume.VolumeMount")
+    @patch("sky.models.VolumeConfig")
+    def test_volume_mounts_with_aws_infra(
+        self, mock_volume_config, mock_volume_mount, mock_task, mock_skypilot_imports
+    ):
+        """Test volume_mounts with AWS infra format (cloud/region/zone)."""
+        mock_task_instance = MagicMock()
+        mock_task.return_value = mock_task_instance
+
+        volume_mounts = [
+            {"path": "/data", "volume_name": "test-vol", "size": "10Gi", "type": "gp2"}
+        ]
+
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            infra="aws/us-east-1/us-east-1a",
+            volume_mounts=volume_mounts,
+        )
+
+        with patch.object(SkypilotExecutor, "to_resources") as mock_to_resources:
+            mock_to_resources.return_value = MagicMock()
+
+            with tempfile.TemporaryDirectory() as tmp_dir:
+                executor.job_dir = tmp_dir
+                executor.to_task("test_task", ["python", "script.py"])
+
+                # Verify AWS infra parsing was covered
+                mock_volume_mount.assert_called_once()
+                mock_volume_config.assert_called_once()
+
+    @patch("sky.task.Task")
+    @patch("sky.volumes.volume.VolumeMount")
+    @patch("sky.models.VolumeConfig")
+    def test_volume_mounts_with_gcp_infra_wildcard(
+        self, mock_volume_config, mock_volume_mount, mock_task, mock_skypilot_imports
+    ):
+        """Test volume_mounts with GCP infra including wildcard zone."""
+        mock_task_instance = MagicMock()
+        mock_task.return_value = mock_task_instance
+
+        volume_mounts = [
+            {"path": "/data", "volume_name": "test-vol", "size": "10Gi", "type": "pd-ssd"}
+        ]
+
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            infra="gcp/us-central1/*",  # Wildcard zone should be skipped
+            volume_mounts=volume_mounts,
+        )
+
+        with patch.object(SkypilotExecutor, "to_resources") as mock_to_resources:
+            mock_to_resources.return_value = MagicMock()
+
+            with tempfile.TemporaryDirectory() as tmp_dir:
+                executor.job_dir = tmp_dir
+                executor.to_task("test_task", ["python", "script.py"])
+
+                # Verify wildcard zone handling was covered
+                mock_volume_mount.assert_called_once()
+                mock_volume_config.assert_called_once()
+
+    @patch("sky.task.Task")
+    @patch("sky.volumes.volume.VolumeMount")
+    @patch("sky.models.VolumeConfig")
+    def test_volume_mounts_fallback_individual_params(
+        self, mock_volume_config, mock_volume_mount, mock_task, mock_skypilot_imports
+    ):
+        """Test volume_mounts fallback to individual cloud/region/zone params when infra is None."""
+        mock_task_instance = MagicMock()
+        mock_task.return_value = mock_task_instance
+
+        volume_mounts = [
+            {"path": "/data", "volume_name": "test-vol", "size": "10Gi", "type": "gp2"}
+        ]
+
+        # Test with individual parameters (no infra) - covers fallback path
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            cloud="aws",
+            region="us-west-2",
+            zone="us-west-2a",
+            volume_mounts=volume_mounts,
+        )
+
+        with patch.object(SkypilotExecutor, "to_resources") as mock_to_resources:
+            mock_to_resources.return_value = MagicMock()
+
+            with tempfile.TemporaryDirectory() as tmp_dir:
+                executor.job_dir = tmp_dir
+                executor.to_task("test_task", ["python", "script.py"])
+
+                # Verify fallback to individual params was covered
+                mock_volume_mount.assert_called_once()
+                mock_volume_config.assert_called_once()
+
+    def test_supports_launcher_transform(self, mock_skypilot_imports):
+        """Test that supports_launcher_transform returns True."""
+        executor = SkypilotExecutor(
+            container_image="nvcr.io/nvidia/nemo:latest",
+            cloud="kubernetes",
+            cluster_name="test-cluster",
+        )
+
+        # Test the method returns True
+        assert executor.supports_launcher_transform() is True