skypilot-nightly 1.0.0.dev20250729__py3-none-any.whl → 1.0.0.dev20250731__py3-none-any.whl

sky/__init__.py

@@ -5,7 +5,7 @@ from typing import Optional
 import urllib.request
 
 # Replaced with the current commit when building the wheels.
-_SKYPILOT_COMMIT_SHA = '1e4f892c08e19e9ca7651d3a114a71d954dcd915'
+_SKYPILOT_COMMIT_SHA = '913f9dd651a2c363231eef760675f1dae49270a1'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20250729'
+__version__ = '1.0.0.dev20250731'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/backends/backend_utils.py

@@ -36,6 +36,7 @@ from sky.jobs import utils as managed_job_utils
 from sky.provision import common as provision_common
 from sky.provision import instance_setup
 from sky.provision.kubernetes import utils as kubernetes_utils
+from sky.serve import serve_utils
 from sky.server.requests import requests as requests_lib
 from sky.skylet import constants
 from sky.usage import usage_lib
@@ -2625,7 +2626,9 @@ def is_controller_accessible(
           failed to be connected.
     """
     if (managed_job_utils.is_consolidation_mode() and
-            controller == controller_utils.Controllers.JOBS_CONTROLLER):
+            controller == controller_utils.Controllers.JOBS_CONTROLLER
+       ) or (serve_utils.is_consolidation_mode() and
+             controller == controller_utils.Controllers.SKY_SERVE_CONTROLLER):
         cn = 'local-controller-consolidation'
         return backends.LocalResourcesHandle(
             cluster_name=cn,

sky/backends/cloud_vm_ray_backend.py

@@ -3209,9 +3209,10 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
                     self._open_ports(handle)
 
         # Capture task YAML and command
-        task_config_redacted = None
+        user_specified_task_config = None
         if task is not None:
-            task_config_redacted = task.to_yaml_config(redact_secrets=True)
+            user_specified_task_config = task.to_yaml_config(
+                use_user_specified_yaml=True)
 
         with timeline.Event('backend.provision.post_process'):
             global_user_state.add_or_update_cluster(
@@ -3220,7 +3221,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
                 set(task.resources),
                 ready=True,
                 config_hash=config_hash,
-                task_config=task_config_redacted,
+                task_config=user_specified_task_config,
             )
             usage_lib.messages.usage.update_final_cluster_status(
                 status_lib.ClusterStatus.UP)

sky/catalog/__init__.py

@@ -91,7 +91,7 @@ def list_accelerator_counts(
     region_filter: Optional[str] = None,
     quantity_filter: Optional[int] = None,
     clouds: CloudFilter = None,
-) -> Dict[str, List[int]]:
+) -> Dict[str, List[float]]:
     """Lists all accelerators offered by Sky and available counts.
 
     Returns: A dictionary of canonical accelerator names mapped to a list
@@ -107,12 +107,12 @@ def list_accelerator_counts(
                                   require_price=False)
     if not isinstance(results, list):
         results = [results]
-    accelerator_counts: Dict[str, Set[int]] = collections.defaultdict(set)
+    accelerator_counts: Dict[str, Set[float]] = collections.defaultdict(set)
     for result in results:
         for gpu, items in result.items():
             for item in items:
                 accelerator_counts[gpu].add(item.accelerator_count)
-    ret: Dict[str, List[int]] = {}
+    ret: Dict[str, List[float]] = {}
     for gpu, counts in accelerator_counts.items():
         ret[gpu] = sorted(counts)
     return ret

sky/catalog/aws_catalog.py

@@ -38,14 +38,26 @@ _DEFAULT_INSTANCE_FAMILY = [
     # CPU: Intel Ice Lake 8375C.
     # Memory: 4 GiB RAM per 1 vCPU;
     'm6i',
+    # This is the latest general-purpose instance family as of Jul 2025.
+    # CPU: Intel Sapphire Rapids.
+    # Memory: 4 GiB RAM per 1 vCPU;
+    'm7i',
     # This is the latest memory-optimized instance family as of Mar 2023.
     # CPU: Intel Ice Lake 8375C
     # Memory: 8 GiB RAM per 1 vCPU;
     'r6i',
+    # This is the latest memory-optimized instance family as of Jul 2025.
+    # CPU: Intel Sapphire Rapids.
+    # Memory: 8 GiB RAM per 1 vCPU;
+    'r7i',
     # This is the latest compute-optimized instance family as of Mar 2023.
     # CPU: Intel Ice Lake 8375C
     # Memory: 2 GiB RAM per 1 vCPU;
     'c6i',
+    # This is the latest compute-optimized instance family as of Jul 2025.
+    # CPU: Intel Sapphire Rapids.
+    # Memory: 2 GiB RAM per 1 vCPU;
+    'c7i',
 ]
 _DEFAULT_NUM_VCPUS = 8
 _DEFAULT_MEMORY_CPU_RATIO = 4

sky/catalog/common.py

@@ -51,7 +51,7 @@ class InstanceTypeInfo(NamedTuple):
     cloud: str
     instance_type: Optional[str]
     accelerator_name: str
-    accelerator_count: int
+    accelerator_count: float
     cpu_count: Optional[float]
     device_memory: Optional[float]
     memory: Optional[float]
@@ -631,7 +631,7 @@ def list_accelerators_impl(
         df = df[df['Region'].str.contains(region_filter,
                                           case=case_sensitive,
                                           regex=True)]
-    df['AcceleratorCount'] = df['AcceleratorCount'].astype(int)
+    df['AcceleratorCount'] = df['AcceleratorCount'].astype(float)
     if quantity_filter is not None:
         df = df[df['AcceleratorCount'] == quantity_filter]
     grouped = df.groupby('AcceleratorName')

sky/catalog/data_fetchers/fetch_aws.py

@@ -46,7 +46,7 @@ ALL_REGIONS = [
     'eu-west-1',
     'eu-west-2',
     'eu-south-1',
-    # 'eu-south-2', # no supported AMI
+    'eu-south-2',
     'eu-west-3',
     'eu-north-1',
     'me-south-1',
@@ -78,6 +78,10 @@ PRICING_TABLE_URL_FMT = 'https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws
 # the permission to query the offerings of the instance.
 # Ref: https://aws.amazon.com/ec2/instance-types/p4/
 P4DE_REGIONS = ['us-east-1', 'us-west-2']
+# g6f instances have fractional GPUs, but the API returns Count: 1 under
+# GpuInfo. However, the GPU memory is properly scaled. Taking the instance GPU
+# divided by the total memory of an L4 will give us the fraction of the GPU.
+L4_GPU_MEMORY = 22888
 
 regions_enabled: Optional[Set[str]] = None
 
@@ -313,6 +317,14 @@ def _get_instance_types_df(region: str) -> Union[str, 'pd.DataFrame']:
                 # AWS API is 'NVIDIA', which is incorrect. See #4652.
                 acc_name = 'H200'
                 acc_count = 8
+            if (row['InstanceType'].startswith('g6f') or
+                    row['InstanceType'].startswith('gr6f')):
+                # These instance actually have only fractional GPUs, but the API
+                # returns Count: 1 under GpuInfo. We need to check the GPU
+                # memory to get the actual fraction of the GPU.
+                # See also Standard_NV{vcpu}ads_A10_v5 support on Azure.
+                fraction = row['GpuInfo']['TotalGpuMemoryInMiB'] / L4_GPU_MEMORY
+                acc_count = round(fraction, 3)
             return pd.Series({
                 'AcceleratorName': acc_name,
                 'AcceleratorCount': acc_count,