gpustack-runtime 0.1.39.post2__tar.gz → 0.1.40__tar.gz

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/.gitignore

@@ -4,9 +4,6 @@
 .DS_Store
 *.swp
 
-# C extensions
-*.so
-
 # Python-generated files
 _version.py
 _version_appendix.py

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.1.39.post2
+Version: 0.1.40
 Summary: GPUStack Runtime is library for detecting GPU resources and launching GPU workloads.
 Project-URL: Homepage, https://github.com/gpustack/runtime
 Project-URL: Bug Tracker, https://github.com/gpustack/gpustack/issues
@@ -15,8 +15,9 @@ Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.10
 Requires-Dist: argcomplete>=3.6.3
 Requires-Dist: docker>=7.1.0
-Requires-Dist: gpustack-runner>=0.1.23.post4
+Requires-Dist: gpustack-runner>=0.1.24.post1
 Requires-Dist: kubernetes>=33.1.0
+Requires-Dist: mthreads-ml-py>=2.2.10
 Requires-Dist: nvidia-ml-py>=13.580.65
 Requires-Dist: podman==5.6.0
 Requires-Dist: pyyaml
@@ -37,6 +38,7 @@ It supports detection of a wide range of GPU and accelerator resources, includin
 - MetaX GPU
 - Moore Threads GPU
 - NVIDIA GPU
+- T-Head PPU
 
 Contributions to support additional GPU resources are welcome!
 

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/README.md

@@ -12,6 +12,7 @@ It supports detection of a wide range of GPU and accelerator resources, includin
 - MetaX GPU
 - Moore Threads GPU
 - NVIDIA GPU
+- T-Head PPU
 
 Contributions to support additional GPU resources are welcome!
 

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/deploy/manifests/docker-compose.yaml

@@ -32,12 +32,15 @@ services:
   #    see https://github.com/gpustack/runtime/blob/016531084000df8075d53772a3c33f41d829ba37/gpustack_runtime/envs.py#L48-L55 for details.
   # - GPUSTACK_RUNTIME_DEPLOY_MIRRORED_NAME:
   #    Specify the name of this Container for gpustack-runtime identify itself.
+  # - GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY:
+  #    Specify the directory of CDI device interface specs.
   #
     container_name: gpustack-runtime
     environment:
       - GPUSTACK_RUNTIME_DEPLOY=Docker
       - GPUSTACK_RUNTIME_DEPLOY_MIRRORED_DEPLOYMENT=true
       - GPUSTACK_RUNTIME_DEPLOY_MIRRORED_NAME=gpustack-runtime
+      - GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY=/var/run/cdi
 
   # Mount necessary volume for gpustack-runtime to work properly.
   # - Mount docker socket for docker runtime.
@@ -47,6 +50,7 @@ services:
     volumes:
       - gpustack-runtime-data:/var/lib/gpustack
       - /var/run/docker.sock:/var/run/docker.sock:ro
+      - /var/run/cdi:/var/run/cdi
   #   - /opt/...:/opt/...:ro
 volumes:
   gpustack-runtime-data: { }

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/deploy/manifests/kubernetes.yaml

@@ -109,6 +109,8 @@ spec:
     #     Change this if you want to deploy workloads into another namespace.
     # - GPUSTACK_RUNTIME_KUBERNETES_NODE_NAME:
     #     Make sure that deploy workloads on the same node.
+    # - GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY:
+    #    Specify the directory of CDI device interface specs.
     #
           env:
             - name: GPUSTACK_RUNTIME_DEPLOY
@@ -127,6 +129,8 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: spec.nodeName
+            - name: GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY
+              value: "/var/run/cdi"
 
     # Mount necessary volume for gpustack-runtime to work properly.
     # - Mount gpustack data volume for persistent data storage.
@@ -135,14 +139,19 @@ spec:
           volumeMounts:
             - name: gpustack-runtime-data
               mountPath: /var/lib/gpustack
+            - name: cdi
+              mountPath: /var/run/cdi
     #       - name: toolkit
     #         mountPath: /opt/...
       volumes:
         - name: gpustack-runtime-data
           emptyDir: { }
+        - name: cdi
+          hostPath:
+            path: /var/run/cdi
     #   - name: toolkit
     #     hostPath:
-    #     name: /opt/...
+    #       path: /opt/...
 
     # Inject service account to the Pod for accessing Kubernetes API.
     #

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/__main__.py

@@ -7,10 +7,12 @@ import time
 from argparse import ArgumentParser
 
 import argcomplete
+from gpustack_runner.cmds import LoadImagesSubCommand
 
 from . import deployer, detector
 from ._version import commit_id, version
 from .cmds import (
+    CDIGenerateSubCommand,
     CopyImagesSubCommand,
     CreateWorkloadSubCommand,
     DeleteWorkloadsSubCommand,
@@ -71,12 +73,14 @@ def main():
     InspectWorkloadSubCommand.register(subcommand_parser)
     DetectDevicesSubCommand.register(subcommand_parser)
     GetDevicesTopologySubCommand.register(subcommand_parser)
-    ListImagesSubCommand.register(subcommand_parser)
-    SaveImagesSubCommand.register(subcommand_parser)
-    CopyImagesSubCommand.register(subcommand_parser)
     LogsSelfSubCommand.register(subcommand_parser)
     ExecSelfSubCommand.register(subcommand_parser)
     InspectSelfSubCommand.register(subcommand_parser)
+    CDIGenerateSubCommand.register(subcommand_parser)
+    ListImagesSubCommand.register(subcommand_parser)
+    SaveImagesSubCommand.register(subcommand_parser)
+    LoadImagesSubCommand.register(subcommand_parser)
+    CopyImagesSubCommand.register(subcommand_parser)
 
     # Autocomplete
     argcomplete.autocomplete(parser)

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.1.39.post2'
-__version_tuple__ = version_tuple = (0, 1, 39, 'post2')
+__version__ = version = '0.1.40'
+__version_tuple__ = version_tuple = (0, 1, 40)
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runtime-0.1.40/gpustack_runtime/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "1f4627e"

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/cmds/__init__.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 from .deployer import (
+    CDIGenerateSubCommand,
     CreateWorkloadSubCommand,
     DeleteWorkloadsSubCommand,
     DeleteWorkloadSubCommand,
@@ -24,6 +25,7 @@ from .images import (
 )
 
 __all__ = [
+    "CDIGenerateSubCommand",
     "CopyImagesSubCommand",
     "CreateWorkloadSubCommand",
     "DeleteWorkloadSubCommand",

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/cmds/deployer.py

@@ -5,7 +5,7 @@ import json
 import os
 import sys
 import time
-from argparse import REMAINDER
+from argparse import OPTIONAL, REMAINDER
 from pathlib import Path
 from typing import TYPE_CHECKING
 
@@ -25,6 +25,7 @@ from ..deployer import (
     WorkloadStatusStateEnum,
     async_logs_self,
     async_logs_workload,
+    cdi_generate_config,
     create_workload,
     delete_workload,
     exec_self,
@@ -34,7 +35,7 @@ from ..deployer import (
     inspect_workload,
     list_workloads,
 )
-from ..detector import supported_backends
+from ..detector import supported_backends, supported_manufacturers
 from .__types__ import SubCommand
 
 if TYPE_CHECKING:
@@ -92,6 +93,7 @@ class CreateWorkloadSubCommand(SubCommand):
     command_script: str | None
     port: int
     host_network: bool
+    privileged: bool
     check: bool
     namespace: str
     name: str
@@ -133,11 +135,19 @@ class CreateWorkloadSubCommand(SubCommand):
 
         deploy_parser.add_argument(
             "--host-network",
+            "--network-host",
             action="store_true",
             help="Use host network (default: False)",
             default=False,
         )
 
+        deploy_parser.add_argument(
+            "--privileged",
+            action="store_true",
+            help="Run the container in privileged mode (default: False)",
+            default=False,
+        )
+
         deploy_parser.add_argument(
             "--check",
             action="store_true",
@@ -183,6 +193,7 @@ class CreateWorkloadSubCommand(SubCommand):
         self.command_script = None
         self.port = args.port
         self.host_network = args.host_network
+        self.privileged = args.privileged
         self.check = args.check
         self.namespace = args.namespace
         self.name = args.name
@@ -237,6 +248,7 @@ class CreateWorkloadSubCommand(SubCommand):
         execution = ContainerExecution(
             command_script=self.command_script,
             args=self.extra_args,
+            privileged=self.privileged,
         )
         ports = (
             [
@@ -945,6 +957,76 @@ class InspectSelfSubCommand(SubCommand):
         print(inspect_self())
 
 
+class CDIGenerateSubCommand(SubCommand):
+    """
+    Command to generate CDI configurations.
+    """
+
+    format: str
+    output: Path | None
+
+    @staticmethod
+    def register(parser: _SubParsersAction):
+        cdi_parser = parser.add_parser(
+            "cdi-generate",
+            help="Generate CDI configurations according to the current environment",
+            aliases=["cdi-gen"],
+        )
+
+        cdi_parser.add_argument(
+            "--format",
+            type=str,
+            choices=["yaml", "json"],
+            default="yaml",
+            help="Format of the CDI configurations",
+        )
+
+        cdi_parser.add_argument(
+            "output",
+            nargs=OPTIONAL,
+            help="Output directory to save CDI configurations (default: current directory)",
+        )
+
+        cdi_parser.set_defaults(func=CDIGenerateSubCommand)
+
+    def __init__(self, args: Namespace):
+        self.format = args.format
+        self.output = Path(args.output) if args.output else None
+
+        if self.output:
+            try:
+                if not self.output.exists():
+                    self.output.mkdir(parents=True, exist_ok=True)
+            except OSError as e:
+                msg = f"Failed to prepare output directory '{self.output}' for CDI configurations"
+                raise RuntimeError(msg) from e
+
+            if not self.output.is_dir():
+                msg = f"The output path '{self.output}' is not a directory"
+                raise RuntimeError(msg)
+
+    def run(self):
+        print("\033[2J\033[H", end="")
+
+        generated = False
+        for manu in supported_manufacturers():
+            content, path = cdi_generate_config(
+                manufacturer=manu,
+                output=self.output,
+            )
+            if content:
+                generated = True
+                if path:
+                    print(f"Generated CDI configuration for '{manu}' at {path}:\n")
+                else:
+                    print(f"Generated CDI configuration for '{manu}':\n")
+                print(content)
+                print()
+
+        if not generated:
+            print("No CDI configurations were generated.")
+
+
 def format_workloads_json(sts: list[WorkloadStatus]) -> str:
     return json.dumps([st.to_dict() for st in sts], indent=2)
 

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/cmds/images.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 from gpustack_runner.cmds import (
     CopyImagesSubCommand,
     ListImagesSubCommand,
+    LoadImagesSubCommand,
     PlatformedImage,
     SaveImagesSubCommand,
     append_images,
@@ -20,6 +21,7 @@ append_images(
 __all__ = [
     "CopyImagesSubCommand",
     "ListImagesSubCommand",
+    "LoadImagesSubCommand",
     "PlatformedImage",
     "SaveImagesSubCommand",
     "append_images",

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/deployer/__init__.py

@@ -31,6 +31,7 @@ from .__types__ import (
     WorkloadStatus,
     WorkloadStatusStateEnum,
 )
+from .cdi import generate_config as cdi_generate_config
 from .docker import (
     DockerDeployer,
     DockerWorkloadPlan,
@@ -602,6 +603,7 @@ __all__ = [
     "WorkloadStatusStateEnum",
     "async_logs_self",
     "async_logs_workload",
+    "cdi_generate_config",
     "create_workload",
     "delete_workload",
     "exec_self",

{gpustack_runtime-0.1.39.post2 → gpustack_runtime-0.1.40}/gpustack_runtime/deployer/__types__.py

@@ -14,6 +14,7 @@ from dataclasses_json import dataclass_json
 
 from .. import envs
 from ..detector import (
+    ManufacturerEnum,
     Topology,
     detect_devices,
     get_devices_topologies,
@@ -21,6 +22,7 @@ from ..detector import (
     manufacturer_to_backend,
 )
 from .__utils__ import (
+    adjust_image_with_envs,
     correct_runner_image,
     fnv1a_32_hex,
     fnv1a_64_hex,
@@ -1018,20 +1020,8 @@ class WorkloadPlan(WorkloadSecurity):
                 c.files.append(command_script)
                 c.execution.command = [command_script_name]  # Override command.
                 c.execution.command_script = None
-            # Add default registry if needed.
-            if (
-                envs.GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY
-                and envs.GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY
-                not in ["docker.io", "index.docker.io"]
-            ):
-                image_registry = (
-                    envs.GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_NAMESPACE
-                )
-                image_split = c.image.split("/")
-                if len(image_split) == 1:
-                    c.image = f"{image_registry}/library/{c.image}"
-                elif len(image_split) == 2:
-                    c.image = f"{image_registry}/{c.image}"
+            # Adjust images.
+            c.image = adjust_image_with_envs(c.image)
             # Correct runner image if needed.
             if envs.GPUSTACK_RUNTIME_DEPLOY_CORRECT_RUNNER_IMAGE:
                 c.image, ok = correct_runner_image(c.image)
@@ -1279,6 +1269,17 @@ class Deployer(ABC):
     """
     Thread pool for the deployer.
     """
+    _visible_devices_manufacturers: dict[str, ManufacturerEnum] | None = None
+    """
+    Recorded visible devices manufacturers,
+    the key is the runtime visible devices env name,
+    the value is the corresponding manufacturer.
+    For example:
+    {
+        "NVIDIA_VISIBLE_DEVICES": ManufacturerEnum.NVIDIA,
+        "AMD_VISIBLE_DEVICES": ManufacturerEnum.AMD
+    }.
+    """
     _visible_devices_env: dict[str, list[str]] | None = None
     """
     Recorded visible devices envs,
@@ -1358,14 +1359,16 @@ class Deployer(ABC):
 
     def _prepare(self):
         """
-        Detect devices once, and construct critical elements for post processing, including:
+        Detect devices once, and construct critical elements for post-processing, including:
+        - Prepare visible devices manufacturers mapping.
         - Prepare visible devices environment variables mapping.
         - Prepare visible devices values mapping.
-        - Prepare topology.
+        - Prepare visible devices topologies mapping.
         """
-        if self._visible_devices_env:
+        if self._visible_devices_manufacturers is not None:
             return
 
+        self._visible_devices_manufacturers = {}
         self._visible_devices_env = {}
         self._visible_devices_cdis = {}
         self._visible_devices_values = {}
@@ -1379,15 +1382,19 @@ class Deployer(ABC):
         if group_devices:
             for manu, devs in group_devices.items():
                 backend = manufacturer_to_backend(manu)
-                rk = envs.GPUSTACK_RUNTIME_DETECT_BACKEND_MAP_RESOURCE_KEY.get(backend)
+                resource_key = (
+                    envs.GPUSTACK_RUNTIME_DETECT_BACKEND_MAP_RESOURCE_KEY.get(backend)
+                )
+                if resource_key is None:
+                    continue
                 ren = envs.GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_RUNTIME_VISIBLE_DEVICES.get(
-                    rk,
+                    resource_key,
                 )
                 ben_list = envs.GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_BACKEND_VISIBLE_DEVICES.get(
-                    rk,
+                    resource_key,
                 )
-                cdi = envs.GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_CONTAINER_DEVICE_INTERFACES.get(
-                    rk,
+                cdi = envs.GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_CDI.get(
+                    resource_key,
                 )
                 if ren and ben_list:
                     valued_uuid = (
@@ -1401,6 +1408,8 @@ class Deployer(ABC):
                         dev_uuids.append(dev.uuid)
                         dev_indexes.append(str(dev.index))
                         dev_indexes_alignment[str(dev.index)] = str(dev_i)
+                    # Map runtime visible devices env <-> manufacturer.
+                    self._visible_devices_manufacturers[ren] = manu
                     # Map runtime visible devices env <-> backend visible devices env list.
                     self._visible_devices_env[ren] = ben_list
                     # Map runtime visible devices env <-> CDI key.
@@ -1433,16 +1442,28 @@ class Deployer(ABC):
                 return
 
         # Fallback to unknown backend
-        self._visible_devices_env["UNKNOWN_RUNTIME_VISIBLE_DEVICES"] = []
-        self._visible_devices_values["UNKNOWN_RUNTIME_VISIBLE_DEVICES"] = ["all"]
+        ren = "UNKNOWN_RUNTIME_VISIBLE_DEVICES"
+        self._visible_devices_manufacturers[ren] = ManufacturerEnum.UNKNOWN
+        self._visible_devices_env[ren] = []
+        self._visible_devices_cdis[ren] = "unknown/devices"
+        self._visible_devices_values[ren] = ["all"]
 
-    def get_visible_devices_values(
+    def get_visible_devices_materials(
         self,
-    ) -> (dict[str, list[str]], dict[str, str], dict[str, list[str]]):
+    ) -> (
+        dict[str, ManufacturerEnum],
+        dict[str, list[str]],
+        dict[str, str],
+        dict[str, list[str]],
+    ):
         """
         Return the visible devices environment variables, cdis and values mappings.
         For example:
         (
+            {
+                "NVIDIA_VISIBLE_DEVICES": ManufacturerEnum.NVIDIA,
+                "AMD_VISIBLE_DEVICES": ManufacturerEnum.AMD
+            },
             {
                 "NVIDIA_VISIBLE_DEVICES": ["CUDA_VISIBLE_DEVICES"],
                 "AMD_VISIBLE_DEVICES": ["HIP_VISIBLE_DEVICES", "ROCR_VISIBLE_DEVICES"]
@@ -1458,10 +1479,12 @@ class Deployer(ABC):
         ).
 
         Returns:
-            A tuple of two dictionaries:
+            A tuple of four dictionaries:
             - The first dictionary maps runtime visible devices environment variable names
-              to lists of backend visible devices environment variable names.
+              to corresponding manufacturers.
             - The second dictionary maps runtime visible devices environment variable names
+              to lists of backend visible devices environment variable names.
+            - The third dictionary maps runtime visible devices environment variable names
               to corresponding CDI keys.
             - The last dictionary maps runtime visible devices environment variable names
               to lists of device indexes or UUIDs.
@@ -1469,6 +1492,7 @@ class Deployer(ABC):
         """
         self._prepare()
         return (
+            self._visible_devices_manufacturers,
             self._visible_devices_env,
             self._visible_devices_cdis,
             self._visible_devices_values,