modal 0.73.6__py3-none-any.whl → 0.73.8__py3-none-any.whl

modal/_functions.py

@@ -476,6 +476,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
         _experimental_buffer_containers: Optional[int] = None,
         _experimental_proxy_ip: Optional[str] = None,
         _experimental_custom_scaling_factor: Optional[float] = None,
+        _experimental_enable_gpu_snapshot: bool = False,
     ) -> "_Function":
         """mdmd:hidden"""
         # Needed to avoid circular imports
@@ -824,6 +825,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
                     _experimental_buffer_containers=_experimental_buffer_containers or 0,
                     _experimental_proxy_ip=_experimental_proxy_ip,
                     _experimental_custom_scaling=_experimental_custom_scaling_factor is not None,
+                    _experimental_enable_gpu_snapshot=_experimental_enable_gpu_snapshot,
                 )
 
                 if isinstance(gpu, list):
@@ -851,6 +853,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
                         _experimental_group_size=function_definition._experimental_group_size,
                         _experimental_buffer_containers=function_definition._experimental_buffer_containers,
                         _experimental_custom_scaling=function_definition._experimental_custom_scaling,
+                        _experimental_enable_gpu_snapshot=_experimental_enable_gpu_snapshot,
                         _experimental_proxy_ip=function_definition._experimental_proxy_ip,
                         snapshot_debug=function_definition.snapshot_debug,
                         runtime_perf_record=function_definition.runtime_perf_record,

modal/_runtime/container_io_manager.py

@@ -491,7 +491,7 @@ class _ContainerIOManager:
         function_call_id: str,
         start_index: int,
         data_format: int,
-        messages_bytes: list[Any],
+        serialized_messages: list[Any],
     ) -> None:
         """Put data onto the `data_out` stream of a function call.
 
@@ -500,7 +500,7 @@ class _ContainerIOManager:
         still use the previous Postgres-backed system based on `FunctionPutOutputs()`.
         """
         data_chunks: list[api_pb2.DataChunk] = []
-        for i, message_bytes in enumerate(messages_bytes):
+        for i, message_bytes in enumerate(serialized_messages):
             chunk = api_pb2.DataChunk(data_format=data_format, index=start_index + i)  # type: ignore
             if len(message_bytes) > MAX_OBJECT_SIZE_BYTES:
                 chunk.data_blob_id = await blob_upload(message_bytes, self._client.stub)
@@ -523,8 +523,8 @@ class _ContainerIOManager:
             # If we don't sleep here for 1ms we end up with an extra call to .put_data_out().
             if index == 1:
                 await asyncio.sleep(0.001)
-            messages_bytes = [serialize_data_format(message, data_format)]
-            total_size = len(messages_bytes[0]) + 512
+            serialized_messages = [serialize_data_format(message, data_format)]
+            total_size = len(serialized_messages[0]) + 512
             while total_size < 16 * 1024 * 1024:  # 16 MiB, maximum size in a single message
                 try:
                     message = message_rx.get_nowait()
@@ -534,10 +534,10 @@ class _ContainerIOManager:
                     received_sentinel = True
                     break
                 else:
-                    messages_bytes.append(serialize_data_format(message, data_format))
-                    total_size += len(messages_bytes[-1]) + 512  # 512 bytes for estimated framing overhead
-            await self.put_data_out(function_call_id, index, data_format, messages_bytes)
-            index += len(messages_bytes)
+                    serialized_messages.append(serialize_data_format(message, data_format))
+                    total_size += len(serialized_messages[-1]) + 512  # 512 bytes for estimated framing overhead
+            await self.put_data_out(function_call_id, index, data_format, serialized_messages)
+            index += len(serialized_messages)
 
     async def _queue_create(self, size: int) -> asyncio.Queue:
         """Create a queue, on the synchronicity event loop (needed on Python 3.8 and 3.9)."""

modal/app.py

@@ -617,6 +617,7 @@ class _App:
         _experimental_buffer_containers: Optional[int] = None,  # Number of additional, idle containers to keep around.
         _experimental_proxy_ip: Optional[str] = None,  # IP address of proxy
         _experimental_custom_scaling_factor: Optional[float] = None,  # Custom scaling factor
+        _experimental_enable_gpu_snapshot: bool = False,  # Experimentally enable GPU memory snapshots.
     ) -> _FunctionDecoratorType:
         """Decorator to register a new Modal [Function](/docs/reference/modal.Function) with this App."""
         if isinstance(_warn_parentheses_missing, _Image):
@@ -759,6 +760,7 @@ class _App:
                 i6pn_enabled=i6pn_enabled,
                 cluster_size=cluster_size,  # Experimental: Clustered functions
                 include_source=include_source if include_source is not None else self._include_source_default,
+                _experimental_enable_gpu_snapshot=_experimental_enable_gpu_snapshot,
             )
 
             self._add_function(function, webhook_config is not None)
@@ -816,6 +818,7 @@ class _App:
         _experimental_buffer_containers: Optional[int] = None,  # Number of additional, idle containers to keep around.
         _experimental_proxy_ip: Optional[str] = None,  # IP address of proxy
         _experimental_custom_scaling_factor: Optional[float] = None,  # Custom scaling factor
+        _experimental_enable_gpu_snapshot: bool = False,  # Experimentally enable GPU memory snapshots.
     ) -> Callable[[CLS_T], CLS_T]:
         """
         Decorator to register a new Modal [Cls](/docs/reference/modal.Cls) with this App.
@@ -890,6 +893,7 @@ class _App:
                 _experimental_buffer_containers=_experimental_buffer_containers,
                 _experimental_proxy_ip=_experimental_proxy_ip,
                 _experimental_custom_scaling_factor=_experimental_custom_scaling_factor,
+                _experimental_enable_gpu_snapshot=_experimental_enable_gpu_snapshot,
             )
 
             self._add_function(cls_func, is_web_endpoint=False)

modal/app.pyi

@@ -197,6 +197,7 @@ class _App:
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_proxy_ip: typing.Optional[str] = None,
         _experimental_custom_scaling_factor: typing.Optional[float] = None,
+        _experimental_enable_gpu_snapshot: bool = False,
     ) -> _FunctionDecoratorType: ...
     @typing_extensions.dataclass_transform(
         field_specifiers=(modal.cls.parameter,),
@@ -241,6 +242,7 @@ class _App:
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_proxy_ip: typing.Optional[str] = None,
         _experimental_custom_scaling_factor: typing.Optional[float] = None,
+        _experimental_enable_gpu_snapshot: bool = False,
     ) -> collections.abc.Callable[[CLS_T], CLS_T]: ...
     async def spawn_sandbox(
         self,
@@ -431,6 +433,7 @@ class App:
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_proxy_ip: typing.Optional[str] = None,
         _experimental_custom_scaling_factor: typing.Optional[float] = None,
+        _experimental_enable_gpu_snapshot: bool = False,
     ) -> _FunctionDecoratorType: ...
     @typing_extensions.dataclass_transform(
         field_specifiers=(modal.cls.parameter,),
@@ -475,6 +478,7 @@ class App:
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_proxy_ip: typing.Optional[str] = None,
         _experimental_custom_scaling_factor: typing.Optional[float] = None,
+        _experimental_enable_gpu_snapshot: bool = False,
     ) -> collections.abc.Callable[[CLS_T], CLS_T]: ...
 
     class __spawn_sandbox_spec(typing_extensions.Protocol[SUPERSELF]):

modal/client.pyi

@@ -27,7 +27,7 @@ class _Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.6"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.8"
     ): ...
     def is_closed(self) -> bool: ...
     @property
@@ -85,7 +85,7 @@ class Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.6"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.8"
     ): ...
     def is_closed(self) -> bool: ...
     @property

modal/functions.pyi

@@ -102,6 +102,7 @@ class Function(
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_proxy_ip: typing.Optional[str] = None,
         _experimental_custom_scaling_factor: typing.Optional[float] = None,
+        _experimental_enable_gpu_snapshot: bool = False,
     ) -> Function: ...
     def _bind_parameters(
         self,

{modal-0.73.6.dist-info → modal-0.73.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: modal
-Version: 0.73.6
+Version: 0.73.8
 Summary: Python client library for Modal
 Author: Modal Labs
 Author-email: support@modal.com

{modal-0.73.6.dist-info → modal-0.73.8.dist-info}/RECORD

@@ -3,7 +3,7 @@ modal/__main__.py,sha256=scYhGFqh8OJcVDo-VOxIT6CCwxOgzgflYWMnIZiMRqE,2871
 modal/_clustered_functions.py,sha256=kTf-9YBXY88NutC1akI-gCbvf01RhMPCw-zoOI_YIUE,2700
 modal/_clustered_functions.pyi,sha256=vllkegc99A0jrUOWa8mdlSbdp6uz36TsHhGxysAOpaQ,771
 modal/_container_entrypoint.py,sha256=qahIuJvaMmWG85N5vNS1yuAQ9XZoo1ftzfatkos_q7I,29553
-modal/_functions.py,sha256=4-CRh1Hgi508tEZRFaSRYdS1Oki0MPgAhnUYcEckMe4,70113
+modal/_functions.py,sha256=XTcpMIlGqa3MPSTbnCwvbagTCeIHlnV4g8WlkGF5SOs,70352
 modal/_ipython.py,sha256=TW1fkVOmZL3YYqdS2YlM1hqpf654Yf8ZyybHdBnlhSw,301
 modal/_location.py,sha256=S3lSxIU3h9HkWpkJ3Pwo0pqjIOSB1fjeSgUsY3x7eec,1202
 modal/_object.py,sha256=ItQcsMNkz9Y3kdTsvfNarbW-paJ2qabDyQ7njaqY0XI,11359
@@ -17,11 +17,11 @@ modal/_traceback.py,sha256=IZQzB3fVlUfMHOSyKUgw0H6qv4yHnpyq-XVCNZKfUdA,5023
 modal/_tunnel.py,sha256=zTBxBiuH1O22tS1OliAJdIsSmaZS8PlnifS_6S5z-mk,6320
 modal/_tunnel.pyi,sha256=JmmDYAy9F1FpgJ_hWx0xkom2nTOFQjn4mTPYlU3PFo4,1245
 modal/_watcher.py,sha256=K6LYnlmSGQB4tWWI9JADv-tvSvQ1j522FwT71B51CX8,3584
-modal/app.py,sha256=4tHmc1hFAL9uGj9wp-u7AOPp2mWfu4wJ-633ghFePIY,44248
-modal/app.pyi,sha256=ppb3UmJU4oX-Ptd5v_SOQJBP1309qovKaJnH5844pFI,25885
+modal/app.py,sha256=wRygVSrWH8iIqhDAAl2Ww_RAkz8MCJZ0Jt9qYZCF6SA,44626
+modal/app.pyi,sha256=lxiuWzE_OLb3WHg-H7Pek9DGBuCUzZ55P594VhJL5LA,26113
 modal/call_graph.py,sha256=1g2DGcMIJvRy-xKicuf63IVE98gJSnQsr8R_NVMptNc,2581
 modal/client.py,sha256=8SQawr7P1PNUCq1UmJMUQXG2jIo4Nmdcs311XqrNLRE,15276
-modal/client.pyi,sha256=sPu1B0hPiIx00baPmkWSaUgJPpL49fxe1-jyFrktW30,7591
+modal/client.pyi,sha256=kOt17KK6b51zHs9T5vme0cCrLzJu-nw1DTE1kSuxXcQ,7591
 modal/cloud_bucket_mount.py,sha256=YOe9nnvSr4ZbeCn587d7_VhE9IioZYRvF9VYQTQux08,5914
 modal/cloud_bucket_mount.pyi,sha256=30T3K1a89l6wzmEJ_J9iWv9SknoGqaZDx59Xs-ZQcmk,1607
 modal/cls.py,sha256=kNnZrBYVXOhgEXU0rDWk2Hr-bQRrsZkMKDgC-TD_6Bs,31063
@@ -40,7 +40,7 @@ modal/file_io.py,sha256=lcMs_E9Xfm0YX1t9U2wNIBPnqHRxmImqjLW1GHqVmyg,20945
 modal/file_io.pyi,sha256=NTRft1tbPSWf9TlWVeZmTlgB5AZ_Zhu2srWIrWr7brk,9445
 modal/file_pattern_matcher.py,sha256=1cZ4V2wSLiaXqAqStETSwp3bzDD6QZOt6pmmjk3Okz4,6505
 modal/functions.py,sha256=kcNHvqeGBxPI7Cgd57NIBBghkfbeFJzXO44WW0jSmao,325
-modal/functions.pyi,sha256=pStDYd9PnJVvWs57qhKIVnmYYAhvKdYzVNoNU1A-s_Q,14231
+modal/functions.pyi,sha256=mtngzj8VlzMOQATe6muBN5oH_Gw9zGKxMKZ56Z-41kU,14288
 modal/gpu.py,sha256=2qZMNnoMrjU-5Bu7fx68pANUAKTtZq0EWEEeBA9OUVQ,7426
 modal/image.py,sha256=Vjsi7wS9dEcoj-7m7_LmvbK5iqEuFz-SHKl2K-qWcew,90952
 modal/image.pyi,sha256=A5mW2dBguEhmRo815Ax1rBIMXTCriu7PqLMHoUPsez8,26372
@@ -83,7 +83,7 @@ modal/volume.py,sha256=JAWeDvoAG95tMBv-fYIERyHsJPS_X_xGpxRRmYtb6j0,30096
 modal/volume.pyi,sha256=kTsXarphjZILXci84LQy7EyC84eXUs5-7D62IM5q3eE,12491
 modal/_runtime/__init__.py,sha256=MIEP8jhXUeGq_eCjYFcqN5b1bxBM4fdk0VESpjWR0fc,28
 modal/_runtime/asgi.py,sha256=c4hmaMW1pLo-cm7ouriJjieuFm4ZF6D2LMy0638sfOs,22139
-modal/_runtime/container_io_manager.py,sha256=2WmFnErVZyfyw3P6iiv1N6FskxZGsiUHrJe7nJ_-iTs,43161
+modal/_runtime/container_io_manager.py,sha256=QVWMCvJatd2696wsauzXl20psxCYsR0d_CHeS5ceTsU,43201
 modal/_runtime/execution_context.py,sha256=E6ofm6j1POXGPxS841X3V7JU6NheVb8OkQc7JpLq4Kg,2712
 modal/_runtime/telemetry.py,sha256=T1RoAGyjBDr1swiM6pPsGRSITm7LI5FDK18oNXxY08U,5163
 modal/_runtime/user_code_imports.py,sha256=zl_Mq9dsrVF62x3w-iNK1YAhZKYAXeFaGpd4G7AySTc,14746
@@ -153,10 +153,10 @@ modal_global_objects/mounts/__init__.py,sha256=MIEP8jhXUeGq_eCjYFcqN5b1bxBM4fdk0
 modal_global_objects/mounts/modal_client_package.py,sha256=W0E_yShsRojPzWm6LtIQqNVolapdnrZkm2hVEQuZK_4,767
 modal_global_objects/mounts/python_standalone.py,sha256=EsC-hdPtiAPOwgW9emHN6muNUkrJwR8dYxroVArxHxM,1841
 modal_proto/__init__.py,sha256=MIEP8jhXUeGq_eCjYFcqN5b1bxBM4fdk0VESpjWR0fc,28
-modal_proto/api.proto,sha256=Ked8dBpkIRf90TAAF1MQkAsu44l3wjVNq7YVX9plzAs,85124
+modal_proto/api.proto,sha256=_RV_hQIWR_CmanZcjieKll-P9P5qVsRDDS7TtGWL4AU,85264
 modal_proto/api_grpc.py,sha256=FYGqDegM_w_qxdtlxum8k31mDibKoMvmNxv_p9cKdKs,109056
-modal_proto/api_pb2.py,sha256=K2y9GzCtyUb7jg6JdX_wL3oAhK_DwWIQytBbm7lu13s,310989
-modal_proto/api_pb2.pyi,sha256=_A_5wXULI7OS24VVbKLGigc_Z6ityG1Xp4Z0weZPzK4,414818
+modal_proto/api_pb2.py,sha256=7hj3Dqmv3Xb6nEzLVMt-zai1EGqUT5uAU5nbGjnk85A,311104
+modal_proto/api_pb2.pyi,sha256=ScYf3xdzV3TQZZR-uPusjrEHoHIxKCRPseS7e8Ee_EI,415384
 modal_proto/api_pb2_grpc.py,sha256=DNp0Et5i_Ey4dKx_1o1LRtYhyWYyT0NzTcAY4EcHn-c,235765
 modal_proto/api_pb2_grpc.pyi,sha256=RI6tWC3L8EIN4-izFSEGPPJl5Ta0lXPNuHUJaWAr35s,54892
 modal_proto/modal_api_grpc.py,sha256=UG8WJU81afrWPwItWB4Ag64E9EpyREMpBbAVGVEYJiM,14550
@@ -170,10 +170,10 @@ modal_proto/options_pb2_grpc.pyi,sha256=CImmhxHsYnF09iENPoe8S4J-n93jtgUYD2JPAc0y
 modal_proto/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 modal_version/__init__.py,sha256=wiJQ53c-OMs0Xf1UeXOxQ7FwlV1VzIjnX6o-pRYZ_Pk,470
 modal_version/__main__.py,sha256=2FO0yYQQwDTh6udt1h-cBnGd1c4ZyHnHSI4BksxzVac,105
-modal_version/_version_generated.py,sha256=7azKeX1fgq54JQqB_4NQJ1XHMuNths4pNX416I_0x_c,148
-modal-0.73.6.dist-info/LICENSE,sha256=psuoW8kuDP96RQsdhzwOqi6fyWv0ct8CR6Jr7He_P_k,10173
-modal-0.73.6.dist-info/METADATA,sha256=ZX7n8GW67cApJpWw-ikTDQVUDBCPdDjcuRds5WiZngU,2329
-modal-0.73.6.dist-info/WHEEL,sha256=G16H4A3IeoQmnOrYV4ueZGKSjhipXx8zc8nu9FGlvMA,92
-modal-0.73.6.dist-info/entry_points.txt,sha256=An-wYgeEUnm6xzrAP9_NTSTSciYvvEWsMZILtYrvpAI,46
-modal-0.73.6.dist-info/top_level.txt,sha256=1nvYbOSIKcmU50fNrpnQnrrOpj269ei3LzgB6j9xGqg,64
-modal-0.73.6.dist-info/RECORD,,
+modal_version/_version_generated.py,sha256=a8LMamoFRZKpp6SoTp5bDH1HRWGljeWLZesEIHf3WIE,148
+modal-0.73.8.dist-info/LICENSE,sha256=psuoW8kuDP96RQsdhzwOqi6fyWv0ct8CR6Jr7He_P_k,10173
+modal-0.73.8.dist-info/METADATA,sha256=wsn98lBqsvSVbab7C3DqlgtiyUF8vmevqqGNf7Ir6WE,2329
+modal-0.73.8.dist-info/WHEEL,sha256=G16H4A3IeoQmnOrYV4ueZGKSjhipXx8zc8nu9FGlvMA,92
+modal-0.73.8.dist-info/entry_points.txt,sha256=An-wYgeEUnm6xzrAP9_NTSTSciYvvEWsMZILtYrvpAI,46
+modal-0.73.8.dist-info/top_level.txt,sha256=1nvYbOSIKcmU50fNrpnQnrrOpj269ei3LzgB6j9xGqg,64
+modal-0.73.8.dist-info/RECORD,,

modal_proto/api.proto

@@ -1263,6 +1263,8 @@ message Function {
   bool _experimental_custom_scaling = 76;
 
   string cloud_provider_str = 77;  // Supersedes cloud_provider
+
+  bool _experimental_enable_gpu_snapshot = 78; // Experimental support for GPU snapshotting
 }
 
 message FunctionAsyncInvokeRequest {
@@ -1376,6 +1378,7 @@ message FunctionData {
   uint32 _experimental_group_size = 19;
   uint32 _experimental_buffer_containers = 22;
   bool _experimental_custom_scaling = 23;
+  bool _experimental_enable_gpu_snapshot = 30;
   string worker_id = 7; // for internal debugging use only
 
   uint32 timeout_secs = 8;