gpustack-runtime 0.1.39.post1__py3-none-any.whl → 0.1.39.post3__py3-none-any.whl

gpustack_runtime/detector/nvidia.py

@@ -2,6 +2,8 @@ from __future__ import annotations
 
 import contextlib
 import logging
+import math
+import time
 from _ctypes import byref
 from functools import lru_cache
 from math import ceil
@@ -76,7 +78,7 @@ class NVIDIADetector(Detector):
     def __init__(self):
         super().__init__(ManufacturerEnum.NVIDIA)
 
-    def detect(self) -> Devices | None:
+    def detect(self) -> Devices | None:  # noqa: PLR0915
         """
         Detect NVIDIA GPUs using pynvml.
 
@@ -125,103 +127,110 @@ class NVIDIADetector(Detector):
             for dev_idx in range(dev_count):
                 dev = pynvml.nvmlDeviceGetHandleByIndex(dev_idx)
 
-                dev_index = dev_idx
-                if envs.GPUSTACK_RUNTIME_DETECT_PHYSICAL_INDEX_PRIORITY:
-                    if dev_files is None:
-                        dev_files = get_device_files(pattern=r"nvidia(?P<number>\d+)")
-                    if len(dev_files) >= dev_count:
-                        dev_file = dev_files[dev_idx]
-                        if dev_file.number is not None:
-                            dev_index = dev_file.number
-                dev_uuid = pynvml.nvmlDeviceGetUUID(dev)
-
-                dev_cores = None
-                if not envs.GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL:
-                    with contextlib.suppress(pycuda.CUDAError):
-                        dev_gpudev = pycuda.cuDeviceGet(dev_idx)
-                        dev_cores = pycuda.cuDeviceGetAttribute(
-                            dev_gpudev,
-                            pycuda.CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
-                        )
+                dev_cc_t = pynvml.nvmlDeviceGetCudaComputeCapability(dev)
+                dev_cc = ".".join(map(str, dev_cc_t))
 
-                dev_mem = 0
-                dev_mem_used = 0
+                dev_bdf = None
                 with contextlib.suppress(pynvml.NVMLError):
-                    dev_mem_info = pynvml.nvmlDeviceGetMemoryInfo(dev)
-                    dev_mem = byte_to_mebibyte(  # byte to MiB
-                        dev_mem_info.total,
-                    )
-                    dev_mem_used = byte_to_mebibyte(  # byte to MiB
-                        dev_mem_info.used,
-                    )
-                if dev_mem == 0:
-                    dev_mem, dev_mem_used = get_memory()
+                    dev_pci_info = pynvml.nvmlDeviceGetPciInfo(dev)
+                    dev_bdf = str(dev_pci_info.busIdLegacy).lower()
 
-                dev_cores_util = None
-                with contextlib.suppress(pynvml.NVMLError):
-                    dev_util_rates = pynvml.nvmlDeviceGetUtilizationRates(dev)
-                    dev_cores_util = dev_util_rates.gpu
-                if dev_cores_util is None:
-                    debug_log_warning(
-                        logger,
-                        "Failed to get device %d cores utilization, setting to 0",
-                        dev_index,
-                    )
-                    dev_cores_util = 0
-
-                dev_temp = None
+                dev_mig_mode = pynvml.NVML_DEVICE_MIG_DISABLE
                 with contextlib.suppress(pynvml.NVMLError):
-                    dev_temp = pynvml.nvmlDeviceGetTemperature(
-                        dev,
-                        pynvml.NVML_TEMPERATURE_GPU,
-                    )
+                    dev_mig_mode, _ = pynvml.nvmlDeviceGetMigMode(dev)
 
-                dev_power = None
-                dev_power_used = None
-                with contextlib.suppress(pynvml.NVMLError):
-                    dev_power = pynvml.nvmlDeviceGetPowerManagementDefaultLimit(dev)
-                    dev_power = dev_power // 1000  # mW to W
-                    dev_power_used = (
-                        pynvml.nvmlDeviceGetPowerUsage(dev) // 1000
-                    )  # mW to W
+                # With MIG disabled, treat as a single device.
+                if dev_mig_mode == pynvml.NVML_DEVICE_MIG_DISABLE:
+                    dev_index = dev_idx
+                    if envs.GPUSTACK_RUNTIME_DETECT_PHYSICAL_INDEX_PRIORITY:
+                        if dev_files is None:
+                            dev_files = get_device_files(
+                                pattern=r"nvidia(?P<number>\d+)",
+                            )
+                        if len(dev_files) >= dev_count:
+                            dev_file = dev_files[dev_idx]
+                            if dev_file.number is not None:
+                                dev_index = dev_file.number
 
-                dev_cc_t = pynvml.nvmlDeviceGetCudaComputeCapability(dev)
-                dev_cc = ".".join(map(str, dev_cc_t))
+                    dev_name = pynvml.nvmlDeviceGetName(dev)
 
-                dev_is_vgpu = False
-                dev_pci_info = pynvml.nvmlDeviceGetPciInfo(dev)
-                for addr in [dev_pci_info.busIdLegacy, dev_pci_info.busId]:
-                    if addr in pci_devs:
-                        dev_is_vgpu = _is_vgpu(pci_devs[addr].config)
-                        break
+                    dev_uuid = pynvml.nvmlDeviceGetUUID(dev)
+
+                    dev_cores = None
+                    if not envs.GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL:
+                        with contextlib.suppress(pycuda.CUDAError):
+                            dev_gpudev = pycuda.cuDeviceGet(dev_idx)
+                            dev_cores = pycuda.cuDeviceGetAttribute(
+                                dev_gpudev,
+                                pycuda.CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
+                            )
+
+                    dev_cores_util = _get_sm_util_from_gpm_metrics(dev)
+                    if dev_cores_util is None:
+                        with contextlib.suppress(pynvml.NVMLError):
+                            dev_util_rates = pynvml.nvmlDeviceGetUtilizationRates(dev)
+                            dev_cores_util = dev_util_rates.gpu
+                    if dev_cores_util is None:
+                        debug_log_warning(
+                            logger,
+                            "Failed to get device %d cores utilization, setting to 0",
+                            dev_index,
+                        )
+                        dev_cores_util = 0
 
-                dev_appendix = {
-                    "arch_family": _get_arch_family(dev_cc_t),
-                    "vgpu": dev_is_vgpu,
-                    "bdf": str(dev_pci_info.busIdLegacy).lower(),
-                }
+                    dev_mem = 0
+                    dev_mem_used = 0
+                    with contextlib.suppress(pynvml.NVMLError):
+                        dev_mem_info = pynvml.nvmlDeviceGetMemoryInfo(dev)
+                        dev_mem = byte_to_mebibyte(  # byte to MiB
+                            dev_mem_info.total,
+                        )
+                        dev_mem_used = byte_to_mebibyte(  # byte to MiB
+                            dev_mem_info.used,
+                        )
+                    if dev_mem == 0:
+                        dev_mem, dev_mem_used = get_memory()
 
-                with contextlib.suppress(pynvml.NVMLError):
-                    dev_fabric = pynvml.c_nvmlGpuFabricInfoV_t()
-                    r = pynvml.nvmlDeviceGetGpuFabricInfoV(dev, byref(dev_fabric))
-                    if r != pynvml.NVML_SUCCESS:
-                        dev_fabric = None
-                    if dev_fabric.state != pynvml.NVML_GPU_FABRIC_STATE_COMPLETED:
-                        dev_fabric = None
-                    if dev_fabric:
-                        dev_appendix["fabric_cluster_uuid"] = stringify_uuid(
-                            bytes(dev_fabric.clusterUuid),
+                    dev_temp = None
+                    with contextlib.suppress(pynvml.NVMLError):
+                        dev_temp = pynvml.nvmlDeviceGetTemperature(
+                            dev,
+                            pynvml.NVML_TEMPERATURE_GPU,
                         )
-                        dev_appendix["fabric_clique_id"] = dev_fabric.cliqueId
 
-                dev_mig_mode = pynvml.NVML_DEVICE_MIG_DISABLE
-                with contextlib.suppress(pynvml.NVMLError):
-                    dev_mig_mode, _ = pynvml.nvmlDeviceGetMigMode(dev)
+                    dev_power = None
+                    dev_power_used = None
+                    with contextlib.suppress(pynvml.NVMLError):
+                        dev_power = pynvml.nvmlDeviceGetPowerManagementDefaultLimit(dev)
+                        dev_power = dev_power // 1000  # mW to W
+                        dev_power_used = (
+                            pynvml.nvmlDeviceGetPowerUsage(dev) // 1000
+                        )  # mW to W
+
+                    dev_is_vgpu = False
+                    if dev_bdf and dev_bdf in pci_devs:
+                        dev_is_vgpu = _is_vgpu(pci_devs[dev_bdf].config)
+
+                    dev_appendix = {
+                        "arch_family": _get_arch_family(dev_cc_t),
+                        "vgpu": dev_is_vgpu,
+                    }
+                    if dev_bdf:
+                        dev_appendix["bdf"] = dev_bdf
 
-                # If MIG is not enabled, return the GPU itself.
+                    with contextlib.suppress(pynvml.NVMLError):
+                        dev_fabric = pynvml.c_nvmlGpuFabricInfoV_t()
+                        r = pynvml.nvmlDeviceGetGpuFabricInfoV(dev, byref(dev_fabric))
+                        if r != pynvml.NVML_SUCCESS:
+                            dev_fabric = None
+                        if dev_fabric.state != pynvml.NVML_GPU_FABRIC_STATE_COMPLETED:
+                            dev_fabric = None
+                        if dev_fabric:
+                            dev_appendix["fabric_cluster_uuid"] = stringify_uuid(
+                                bytes(dev_fabric.clusterUuid),
+                            )
+                            dev_appendix["fabric_clique_id"] = dev_fabric.cliqueId
 
-                if dev_mig_mode == pynvml.NVML_DEVICE_MIG_DISABLE:
-                    dev_name = pynvml.nvmlDeviceGetName(dev)
                     ret.append(
                         Device(
                             manufacturer=self.manufacturer,
@@ -283,13 +292,20 @@ class NVIDIADetector(Detector):
                         pynvml.nvmlDeviceGetPowerUsage(mdev) // 1000
                     )  # mW to W
 
-                    mdev_appendix = dev_appendix.copy()
+                    mdev_appendix = {
+                        "arch_family": _get_arch_family(dev_cc_t),
+                        "vgpu": True,
+                    }
+                    if dev_bdf:
+                        mdev_appendix["bdf"] = dev_bdf
 
                     mdev_gi_id = pynvml.nvmlDeviceGetGpuInstanceId(mdev)
                     mdev_appendix["gpu_instance_id"] = mdev_gi_id
                     mdev_ci_id = pynvml.nvmlDeviceGetComputeInstanceId(mdev)
                     mdev_appendix["compute_instance_id"] = mdev_ci_id
 
+                    mdev_cores_util = _get_sm_util_from_gpm_metrics(dev, mdev_gi_id)
+
                     if not mdev_name:
                         mdev_attrs = pynvml.nvmlDeviceGetAttributes(mdev)
 
@@ -374,6 +390,7 @@ class NVIDIADetector(Detector):
                             runtime_version_original=sys_runtime_ver_original,
                             compute_capability=dev_cc,
                             cores=mdev_cores,
+                            cores_utilization=mdev_cores_util,
                             memory=mdev_mem,
                             memory_used=mdev_mem_used,
                             memory_utilization=get_utilization(mdev_mem_used, mdev_mem),
@@ -492,6 +509,97 @@ class NVIDIADetector(Detector):
         return ret
 
 
+def _get_gpm_metrics(
+    metrics: list[int],
+    dev: pynvml.c_nvmlDevice_t,
+    gpu_instance_id: int | None = None,
+    interval: float = 0.1,
+) -> list[pynvml.c_nvmlGpmMetric_t] | None:
+    """
+    Get GPM metrics for a device or a MIG GPU instance.
+
+    Args:
+        metrics:
+            A list of GPM metric IDs to query.
+        dev:
+            The NVML device handle.
+        gpu_instance_id:
+            The GPU instance ID for MIG devices.
+        interval:
+            Interval in seconds between two samples.
+
+    Returns:
+        A list of GPM metric structures, or None if failed.
+
+    """
+    try:
+        dev_gpm_support = pynvml.nvmlGpmQueryDeviceSupport(dev)
+        if not bool(dev_gpm_support.isSupportedDevice):
+            return None
+    except pynvml.NVMLError:
+        debug_log_warning(logger, "Unsupported GPM query")
+        return None
+
+    dev_gpm_metrics = pynvml.c_nvmlGpmMetricsGet_t()
+    try:
+        dev_gpm_metrics.sample1 = pynvml.nvmlGpmSampleAlloc()
+        dev_gpm_metrics.sample2 = pynvml.nvmlGpmSampleAlloc()
+        if gpu_instance_id is None:
+            pynvml.nvmlGpmSampleGet(dev, dev_gpm_metrics.sample1)
+            time.sleep(interval)
+            pynvml.nvmlGpmSampleGet(dev, dev_gpm_metrics.sample2)
+        else:
+            pynvml.nvmlGpmMigSampleGet(dev, gpu_instance_id, dev_gpm_metrics.sample1)
+            time.sleep(interval)
+            pynvml.nvmlGpmMigSampleGet(dev, gpu_instance_id, dev_gpm_metrics.sample2)
+        dev_gpm_metrics.version = pynvml.NVML_GPM_METRICS_GET_VERSION
+        dev_gpm_metrics.numMetrics = len(metrics)
+        for metric_idx, metric in enumerate(metrics):
+            dev_gpm_metrics.metrics[metric_idx].metricId = metric
+        pynvml.nvmlGpmMetricsGet(dev_gpm_metrics)
+    except pynvml.NVMLError:
+        debug_log_exception(logger, "Failed to get GPM metrics")
+        return None
+    finally:
+        if dev_gpm_metrics.sample1:
+            pynvml.nvmlGpmSampleFree(dev_gpm_metrics.sample1)
+        if dev_gpm_metrics.sample2:
+            pynvml.nvmlGpmSampleFree(dev_gpm_metrics.sample2)
+    return list(dev_gpm_metrics.metrics)
+
+
+def _get_sm_util_from_gpm_metrics(
+    dev: pynvml.c_nvmlDevice_t,
+    gpu_instance_id: int | None = None,
+    interval: float = 0.1,
+) -> int | None:
+    """
+    Get SM utilization from GPM metrics.
+
+    Args:
+        dev:
+            The NVML device handle.
+        gpu_instance_id:
+            The GPU instance ID for MIG devices.
+        interval:
+            Interval in seconds between two samples.
+
+    Returns:
+        The SM utilization as an integer percentage, or None if failed.
+
+    """
+    dev_gpm_metrics = _get_gpm_metrics(
+        metrics=[pynvml.NVML_GPM_METRIC_SM_UTIL],
+        dev=dev,
+        gpu_instance_id=gpu_instance_id,
+        interval=interval,
+    )
+    if dev_gpm_metrics and not math.isnan(dev_gpm_metrics[0].value):
+        return int(dev_gpm_metrics[0].value)
+
+    return None
+
+
 def _get_arch_family(dev_cc_t: list[int]) -> str:
     """
     Get the architecture family based on the CUDA compute capability.

gpustack_runtime/detector/pyhsa/__init__.py

@@ -446,13 +446,13 @@ def has_agent_get_asic_family_id(agent):
 
 @dataclass
 class Agent:
-    device_type: int
-    device_id: str
-    uuid: str
-    name: str
-    compute_capability: str
-    compute_units: int
-    asic_family_id: int | None
+    device_type: int | None = None
+    device_id: str | None = None
+    uuid: str | None = None
+    name: str | None = None
+    compute_capability: str | None = None
+    compute_units: int | None = None
+    asic_family_id: int | None = None
 
 
 def get_agents() -> list[Agent]:

gpustack_runtime/detector/pyrocmsmi/__init__.py

@@ -223,15 +223,9 @@ def rsmi_dev_target_graphics_version_get(device=0):
         c_version = c_uint64()
         ret = rocmsmiLib.rsmi_dev_target_graphics_version_get(device, byref(c_version))
         _rocmsmiCheckReturn(ret)
-        version = str(c_version.value)
-        if len(version) == 4:
-            dev_name = rsmi_dev_name_get(device)
-            if "Instinct MI2" in dev_name:
-                hex_part = str(hex(int(version[2:]))).replace("0x", "")
-                version = version[:2] + hex_part
-        else:
-            version = str(c_version.value // 10 + c_version.value % 10)
-        return "gfx" + version
+        if c_version.value < 2000:
+            return "gfx" + str(c_version.value)
+        return "gfx" + hex(c_version.value)[2:]
     except AttributeError:
         return None
 

gpustack_runtime/envs.py

@@ -119,25 +119,25 @@ if TYPE_CHECKING:
     """
     Correct the gpustack-runner image by rendering it with the host's detection.
     """
-    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY: str | None = None
+    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY: str | None = None
     """
     Default container registry for deployer to pull images from.
     If not set, it should be "docker.io".
     If the image name already contains a registry, this setting will be ignored.
     """
-    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_NAMESPACE: str | None = None
-    """
-    Namespace for default runner images.
-    If not set, it should be "gpustack".
-    """
-    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_USERNAME: str | None = None
+    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY_USERNAME: str | None = None
     """
     Username for the default container registry.
     """
-    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_PASSWORD: str | None = None
+    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY_PASSWORD: str | None = None
     """
     Password for the default container registry.
     """
+    GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_NAMESPACE: str | None = None
+    """
+    Namespace for default runtime container images.
+    If not set, it should be "gpustack".
+    """
     GPUSTACK_RUNTIME_DEPLOY_IMAGE_PULL_POLICY: str | None = None
     """
     Image pull policy for the deployer (e.g., Always, IfNotPresent, Never).
@@ -383,36 +383,48 @@ variables: dict[str, Callable[[], Any]] = {
     "GPUSTACK_RUNTIME_DEPLOY_CORRECT_RUNNER_IMAGE": lambda: to_bool(
         getenv("GPUSTACK_RUNTIME_DEPLOY_CORRECT_RUNNER_IMAGE", "1"),
     ),
-    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY": lambda: trim_str(
+    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY": lambda: trim_str(
         getenvs(
             keys=[
-                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY",
+                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY",
                 # TODO(thxCode): Backward compatibility, remove in v0.1.45 later.
                 "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_REGISTRY",
                 # Compatible with gpustack/gpustack.
                 "GPUSTACK_SYSTEM_DEFAULT_CONTAINER_REGISTRY",
+                # Legacy compatibility.
+                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY",
             ],
         ),
     ),
-    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_NAMESPACE": lambda: trim_str(
-        getenv("GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_NAMESPACE"),
-    ),
-    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_USERNAME": lambda: trim_str(
+    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY_USERNAME": lambda: trim_str(
         getenvs(
             keys=[
-                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_USERNAME",
+                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY_USERNAME",
                 # TODO(thxCode): Backward compatibility, remove in v0.1.45 later.
                 "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_REGISTRY_USERNAME",
+                # Legacy compatibility.
+                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_USERNAME",
             ],
         ),
     ),
-    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_PASSWORD": lambda: getenvs(
+    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY_PASSWORD": lambda: getenvs(
         keys=[
-            "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_PASSWORD",
+            "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_REGISTRY_PASSWORD",
             # TODO(thxCode): Backward compatibility, remove in v0.1.45 later.
             "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_REGISTRY_PASSWORD",
+            # Legacy compatibility.
+            "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_REGISTRY_PASSWORD",
         ],
     ),
+    "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_NAMESPACE": lambda: trim_str(
+        getenvs(
+            keys=[
+                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_CONTAINER_NAMESPACE",
+                # Legacy compatibility.
+                "GPUSTACK_RUNTIME_DEPLOY_DEFAULT_IMAGE_NAMESPACE",
+            ],
+        ),
+    ),
     "GPUSTACK_RUNTIME_DEPLOY_IMAGE_PULL_POLICY": lambda: choice(
         getenv(
             "GPUSTACK_RUNTIME_DEPLOY_IMAGE_PULL_POLICY",
@@ -464,7 +476,7 @@ variables: dict[str, Callable[[], Any]] = {
             "hygon.com/devices=HIP_VISIBLE_DEVICES;"
             "iluvatar.ai/devices=CUDA_VISIBLE_DEVICES;"
             "metax-tech.com/devices=CUDA_VISIBLE_DEVICES;"
-            "mthreads.com/devices=CUDA_VISIBLE_DEVICES;"
+            "mthreads.com/devices=CUDA_VISIBLE_DEVICES,MUSA_VISIBLE_DEVICES;"
             "nvidia.com/devices=CUDA_VISIBLE_DEVICES;",
         ),
         list_sep=",",

{gpustack_runtime-0.1.39.post1.dist-info → gpustack_runtime-0.1.39.post3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.1.39.post1
+Version: 0.1.39.post3
 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.post2
+Requires-Dist: gpustack-runner>=0.1.23.post5
 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

{gpustack_runtime-0.1.39.post1.dist-info → gpustack_runtime-0.1.39.post3.dist-info}/RECORD

@@ -1,49 +1,48 @@
 gpustack_runtime/__init__.py,sha256=kKcK6DMIXOdpWhgMS7xlsNrBNvEmY0L8eyQtBIC3CU4,160
-gpustack_runtime/__main__.py,sha256=ILEyXTfxn__jFvovxjxzRDIg7QJqQ2pQrP_2BCGQZRQ,3389
-gpustack_runtime/_version.py,sha256=nC-EU-83isDNnCdVSshJtz4OPykhWYA4jFlaXf9xxAY,792
+gpustack_runtime/__main__.py,sha256=YfxBJy7Xu87jsl1g9EWXtEeAWkLqX9nbHuty7gGUgnw,3620
+gpustack_runtime/_version.py,sha256=j_bFBefu7NnpCEW3cuer6_aWK5fUvN8thDABLTT31c8,792
 gpustack_runtime/_version.pyi,sha256=A42NoSgcqEXVy2OeNm4LXC9CbyonbooYrSUBlPm2lGY,156
-gpustack_runtime/envs.py,sha256=624z3eettjvXY3MW2KXFvPcWogp-gmjI5SoBnyfF9Vs,31852
+gpustack_runtime/envs.py,sha256=1tNIZIzPuIMbBPMy6gUkadlUwmP_6qpMmCZQD_qwRNI,32429
 gpustack_runtime/logging.py,sha256=h_fvD5FV7GHCo00IUDLQmkPR-H6r66IX_WSwZwl0mCw,6869
-gpustack_runtime/cmds/__init__.py,sha256=9licqBPf2qLsGmv_cL6-SSUPVYCLavcRvryFfr_ZHUk,1010
+gpustack_runtime/cmds/__init__.py,sha256=82MqustO4mRDCytBiT4SUMDIOaa_4n81sJEXQwwfqHE,1166
 gpustack_runtime/cmds/__types__.py,sha256=7C4kQM0EHPD8WpJpTo6kh9rEdkrYALcLQ-GAzMMsqV8,789
-gpustack_runtime/cmds/deployer.py,sha256=GlQKEXlrC35M7jFmUGylVlfndIu3Y9FhYTo5SsDMjPo,23206
+gpustack_runtime/cmds/deployer.py,sha256=7tTBf2_pKtK3_jeGcwjxg3ibzUBUKMKyGU5W6Nms7vE,26222
 gpustack_runtime/cmds/detector.py,sha256=xTuPkl0Psi6T_n7RvwCHBeubX4AsjXAhc_emrMufY58,8724
 gpustack_runtime/cmds/images.py,sha256=CjmD82JMdzftRiaeEYSKTGBXudb_qla41v-UzwZQtTo,514
-gpustack_runtime/deployer/__init__.py,sha256=OGAr302rnH8ijmqUPxaF8MjyrDcCdw2BF0wnq_-GKXc,10346
+gpustack_runtime/deployer/__init__.py,sha256=2LQSFBO4syh7Q1rp-Rd_122Vwh-4CNDM1CGxmu3GhXY,15498
 gpustack_runtime/deployer/__patches__.py,sha256=4TCUdDzKBiOYmr5s0UkrEjBU0UjT7U36R0aQbSvrFRE,17823
-gpustack_runtime/deployer/__types__.py,sha256=UWovjd3n1avWwNm7U2z3Ipftpa9_r7ABUKu0PE1sVy4,60531
-gpustack_runtime/deployer/__utils__.py,sha256=aKoHSeodwEPuUW19bXZNNPG-5ACFn-o2EosmGxtkxS0,21054
-gpustack_runtime/deployer/docker.py,sha256=moxCOn_IkdQCiHxZKgOvGGOT3TbemJTq74vhgyBlHZY,77361
-gpustack_runtime/deployer/kuberentes.py,sha256=BSlSwsp0CK0xoSdQf9U4EnbbOei8pIk1QQi3p2lzHz4,79614
-gpustack_runtime/deployer/podman.py,sha256=GUxMTnaQuErZK55wvhK2MbgFjkYLrXD8ThXuWcM06zI,74114
+gpustack_runtime/deployer/__types__.py,sha256=va80QrwaVJx3WJSEIlQvjZYbeuwG6PL04sTXl90q3rg,71365
+gpustack_runtime/deployer/__utils__.py,sha256=wBnyKhR1WLcdPQlhjKeEgJ9kPvBhxVlTAqbW2iBBLIE,21710
+gpustack_runtime/deployer/docker.py,sha256=vrqglLLANiO5P4l0j828KbdZ_-wGnbol0PnB7QUi7qs,84391
+gpustack_runtime/deployer/kuberentes.py,sha256=JR79UfdMCo_0GCpeOFHxDQFAnYZdu7oApNkcjhr3DzA,87333
+gpustack_runtime/deployer/podman.py,sha256=nKjsNxgeAbSuc5Lwbyx4W_P-lY7dAvl6WpjExRT-ENQ,81520
 gpustack_runtime/detector/__init__.py,sha256=kLVxZ9iud2aLwqqAOanIYNb4QSHavUPhqay-FU5ndbU,5443
 gpustack_runtime/detector/__types__.py,sha256=nVh9OV8tZdHFjHPlYul3u2zMvnpa9KOL5GxlPJhEvGk,13163
-gpustack_runtime/detector/__utils__.py,sha256=Yj4GvGvuDuscDG5OpExYdv-1VhmEHBpRrBC9rTsB-kA,24584
-gpustack_runtime/detector/amd.py,sha256=_sRBRZn9pFQ9jITCetvsoHicjXA9RSVrFvo6zOwCHU8,18356
+gpustack_runtime/detector/__utils__.py,sha256=3_6RsBmrFJ0TpNk0KkwTju_cWhXA-rP__ZNktjP_T_Q,25269
+gpustack_runtime/detector/amd.py,sha256=JGERVoSx1MlZP9AHs_u92I_Zz4Cl4-fxMzLMBia_F74,18693
 gpustack_runtime/detector/ascend.py,sha256=VkmNM-enbjMNVXKFfv7z6S-4BkwD6YkBmcC0J1ZGEuc,18588
 gpustack_runtime/detector/cambricon.py,sha256=bqWo2mUuf3a5kV8NC1PzFtHIUhVQkSejUSWR5Z5XaGA,3899
-gpustack_runtime/detector/hygon.py,sha256=I4SWuq5hXLRPJQyIyCoSaU1VnRMDblzQtljKy7CBVdM,13337
-gpustack_runtime/detector/iluvatar.py,sha256=1fbkfE2w_MH36YgQDv21xb76sl6GvYxVd3qSYXjET0g,10372
+gpustack_runtime/detector/hygon.py,sha256=n_spB46BFSQtUBYEact0bQo25uOWSn0fsZ8Uwcq9DHs,13540
+gpustack_runtime/detector/iluvatar.py,sha256=IRc1teRB21b8L0ouCT61VL3eDaWcLWwoKpwFgjHILpg,10693
 gpustack_runtime/detector/metax.py,sha256=YRpZVfz_nG7QTEMqTxqpgi9x5aPYy4gfvYhiUpbz8R8,10146
-gpustack_runtime/detector/mthreads.py,sha256=PoGirid-rboH4Nm6sKnv424ULax1eu1rEaAsWKPT02g,11105
-gpustack_runtime/detector/nvidia.py,sha256=eKVyvyvwbXOIBtWlaCMltqB_ohW9yBnw8M_KYHzv81E,26445
+gpustack_runtime/detector/mthreads.py,sha256=XW8dyb4i8qVKxHw1Yu08f2YRhpj0Xe8E-MSNfSeOwr4,11012
+gpustack_runtime/detector/nvidia.py,sha256=HmctG6s_03FsISeIJ-mptoQ9CUnQyE0X3bh-Qfa4dnI,30306
 gpustack_runtime/detector/pyacl/__init__.py,sha256=13_zyQxo_yql99Ex7n3HTPzZVTSt8A-TN0-u5eDW8nE,15994
 gpustack_runtime/detector/pyamdgpu/__init__.py,sha256=vFIOA072L01Nxvzgixns-wNrQJK8sJMvEs90DEEE2-0,8378
 gpustack_runtime/detector/pyamdsmi/__init__.py,sha256=BXAOVnoew6MliU86ci73Rk2yWPeXRUBicFKFWh-r2s0,929
 gpustack_runtime/detector/pycuda/__init__.py,sha256=o2Oj3l0HsXNttwnUylxerbupK-45gZWbluqfEvBDQ38,16673
 gpustack_runtime/detector/pydcmi/__init__.py,sha256=KVmem7Cd6ZoQ-irB_WOsDVOiyGr5Cv69_rm1NNBpkxo,37129
-gpustack_runtime/detector/pyhsa/__init__.py,sha256=jQ6tOPG8kn5Xp3F99-Vu6aUc4hzSUOdoGkEcl8qxQOc,15354
+gpustack_runtime/detector/pyhsa/__init__.py,sha256=v2rWLyyqNQ24Jpy-7MoBbctd797gHN_6JNrFA84AjxM,15445
 gpustack_runtime/detector/pyixml/__init__.py,sha256=oYFAfQ8ZbPg7oMFXFfLjJuCBEV7hWQpFqMMVcN99Sxs,166564
-gpustack_runtime/detector/pymtml/__init__.py,sha256=hfaVZysit_MUTDs0qsnOBgKXq_8DKlCb0Jv00WasU0w,22008
 gpustack_runtime/detector/pymxsml/__init__.py,sha256=spHMs9t2tiV-C9kO-cV_230CBVAZ-j2DWtLfiFU5V3E,25
 gpustack_runtime/detector/pymxsml/mxsml.py,sha256=nRuKSVxIsJj2SJX-SYnE9NGZ7hGjWO9nKczC6sYmSRg,47196
 gpustack_runtime/detector/pymxsml/mxsml_extension.py,sha256=zfWFRGa9eSup336t2lPhIvCbPpa7Prn-7xCHRljg4CI,26967
 gpustack_runtime/detector/pymxsml/mxsml_mcm.py,sha256=a4rX7hJNJKTqLodKU9rYeDaAEKef6UNVThl1w-aiFsA,15363
 gpustack_runtime/detector/pyrocmcore/__init__.py,sha256=8XxKmwLX4-uoP7cfxjGoEmk9qlrGf0804pgyb74mJKs,2108
-gpustack_runtime/detector/pyrocmsmi/__init__.py,sha256=OmvfJepAtEnDKBMLrktMXlRbJEZttMDQ2R4ztD8kLGg,11806
-gpustack_runtime/_version_appendix.py,sha256=8YQIRWVgq2vdpyqBhE6QHm4WYQetOtYCK_sPA5OF-E0,23
-gpustack_runtime-0.1.39.post1.dist-info/METADATA,sha256=E4bWmJgRHV2dKcb87lEDMQYapW5xoFeFwmtmMnPPIyU,2131
-gpustack_runtime-0.1.39.post1.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-gpustack_runtime-0.1.39.post1.dist-info/entry_points.txt,sha256=bBO_61GxP6dIT74uZwbSDgW5Vt2pTePUS3CgjUJkUgg,68
-gpustack_runtime-0.1.39.post1.dist-info/licenses/LICENSE,sha256=OiPibowBvB-NHV3TP_NOj18XNBlXcshXZFMpa3uvKVE,10362
-gpustack_runtime-0.1.39.post1.dist-info/RECORD,,
+gpustack_runtime/detector/pyrocmsmi/__init__.py,sha256=gtyiS4vXpuc9Y9xt3cw_ObqiivbsU8_xi48TIqZA0hk,11538
+gpustack_runtime/_version_appendix.py,sha256=8EQMM2co0WsEEaFTlEGHr7PFAY9g_8S8iAhM1Hgptog,23
+gpustack_runtime-0.1.39.post3.dist-info/METADATA,sha256=XWfZJrvw3xPJsjfwnFtQoCNv3SR7xBNlFjppk9ErwgE,2169
+gpustack_runtime-0.1.39.post3.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+gpustack_runtime-0.1.39.post3.dist-info/entry_points.txt,sha256=bBO_61GxP6dIT74uZwbSDgW5Vt2pTePUS3CgjUJkUgg,68
+gpustack_runtime-0.1.39.post3.dist-info/licenses/LICENSE,sha256=OiPibowBvB-NHV3TP_NOj18XNBlXcshXZFMpa3uvKVE,10362
+gpustack_runtime-0.1.39.post3.dist-info/RECORD,,