lightning-sdk 0.2.20__py3-none-any.whl → 0.2.21__py3-none-any.whl

lightning_sdk/__init__.py

@@ -31,6 +31,6 @@ __all__ = [
     "User",
 ]
 
-__version__ = "0.2.20"
+__version__ = "0.2.21"
 _check_version_and_prompt_upgrade(__version__)
 _set_tqdm_envvars_noninteractive()

lightning_sdk/api/deployment_api.py

@@ -282,10 +282,11 @@ class DeploymentApi:
 
         requires_release = False
         requires_release |= apply_change(deployment.spec, "image", image)
+
         requires_release |= apply_change(deployment.spec, "entrypoint", entrypoint)
         requires_release |= apply_change(deployment.spec, "command", command)
         requires_release |= apply_change(deployment.spec, "env", to_env(env))
-        requires_release |= apply_change(deployment.spec, "readiness_probe", to_health_check(health_check))
+        requires_release |= apply_change(deployment.spec, "readiness_probe", to_health_check(health_check, False))
         requires_release |= apply_change(deployment.spec, "cluster_id", cloud_account)
         requires_release |= apply_change(deployment.spec, "spot", spot)
         requires_release |= apply_change(deployment.spec, "quantity", quantity)
@@ -524,8 +525,12 @@ def to_endpoint(
 
 
 def to_health_check(
-    health_check: Optional[Union[HttpHealthCheck, ExecHealthCheck]] = None
+    health_check: Optional[Union[HttpHealthCheck, ExecHealthCheck]] = None,
+    use_default: bool = True,
 ) -> Optional[V1JobHealthCheckConfig]:
+    if health_check is None and not use_default:
+        return None
+
     # Use Default health check if none is provided
     if not health_check:
         return V1JobHealthCheckConfig(

lightning_sdk/api/license_api.py

@@ -5,13 +5,13 @@ from urllib.parse import urlencode
 from lightning_sdk.lightning_cloud import env
 from lightning_sdk.lightning_cloud.rest_client import LightningClient
 
-LICENSE_CODE = os.environ.get("LICENSE_CODE", "we843fiji89")
+LICENSE_CODE = os.environ.get("LICENSE_CODE", "d9s79g79ss")
 # https://lightning.ai/home?settings=licenses
-LICENSE_SIGNING_URL = f"{env.LIGHTNING_CLOUD_URL}/ai-hub?settings=licenses"
+LICENSE_SIGNING_URL = f"{env.LIGHTNING_CLOUD_URL}?settings=licenses"
 
 
 def generate_url_user_settings(redirect_to: str = LICENSE_SIGNING_URL) -> str:
-    params = urlencode({"redirectTo": redirect_to, "mode": "licenses", "okbhrt": LICENSE_CODE})
+    params = urlencode({"redirectTo": redirect_to, "okbhrt": LICENSE_CODE})
     return f"{env.LIGHTNING_CLOUD_URL}/sign-in?{params}"
 
 

lightning_sdk/api/llm_api.py

@@ -1,7 +1,10 @@
+import asyncio
 import base64
 import json
 import os
-from typing import Dict, Generator, List, Optional, Union
+import threading
+import warnings
+from typing import Any, AsyncGenerator, Dict, Generator, List, Optional, Union
 
 from pip._vendor.urllib3 import HTTPResponse
 
@@ -27,35 +30,40 @@ class LLMApi:
         result = self._client.assistants_service_list_assistants(user_id=user_id)
         return result.assistants
 
+    def _parse_stream_line(self, decoded_line: str) -> Optional[V1ConversationResponseChunk]:
+        try:
+            payload = json.loads(decoded_line)
+            result_data = payload.get("result", {})
+
+            choices = []
+            for choice in result_data.get("choices", []):
+                delta = choice.get("delta", {})
+                choices.append(
+                    V1ResponseChoice(
+                        delta=V1ResponseChoiceDelta(**delta),
+                        finish_reason=choice.get("finishReason"),
+                        index=choice.get("index"),
+                    )
+                )
+
+            return V1ConversationResponseChunk(
+                choices=choices,
+                conversation_id=result_data.get("conversationId"),
+                executable=result_data.get("executable"),
+                id=result_data.get("id"),
+                throughput=result_data.get("throughput"),
+            )
+        except json.JSONDecodeError:
+            warnings.warn("Error decoding JSON:", decoded_line)
+            return None
+
     def _stream_chat_response(self, result: HTTPResponse) -> Generator[V1ConversationResponseChunk, None, None]:
         for line in result.stream():
             decoded_lines = line.decode("utf-8").strip()
             for decoded_line in decoded_lines.splitlines():
-                try:
-                    payload = json.loads(decoded_line)
-                    result_data = payload.get("result", {})
-
-                    choices = []
-                    for choice in result_data.get("choices", []):
-                        delta = choice.get("delta", {})
-                        choices.append(
-                            V1ResponseChoice(
-                                delta=V1ResponseChoiceDelta(**delta),
-                                finish_reason=choice.get("finishReason"),
-                                index=choice.get("index"),
-                            )
-                        )
-
-                    yield V1ConversationResponseChunk(
-                        choices=choices,
-                        conversation_id=result_data.get("conversationId"),
-                        executable=result_data.get("executable"),
-                        id=result_data.get("id"),
-                        throughput=result_data.get("throughput"),
-                    )
-
-                except json.JSONDecodeError:
-                    print("Error decoding JSON:", decoded_line)
+                chunk = self._parse_stream_line(decoded_line)
+                if chunk:
+                    yield chunk
 
     def _encode_image_bytes_to_data_url(self, image: str) -> str:
         with open(image, "rb") as image_file:
@@ -110,6 +118,90 @@ class LLMApi:
             return result.result
         return self._stream_chat_response(result)
 
+    async def async_start_conversation(
+        self,
+        prompt: str,
+        system_prompt: Optional[str],
+        max_completion_tokens: int,
+        assistant_id: str,
+        images: Optional[List[str]] = None,
+        conversation_id: Optional[str] = None,
+        billing_project_id: Optional[str] = None,
+        name: Optional[str] = None,
+        metadata: Optional[Dict[str, str]] = None,
+        stream: bool = False,
+    ) -> Union[V1ConversationResponseChunk, AsyncGenerator[V1ConversationResponseChunk, None]]:
+        is_internal_conversation = os.getenv("LIGHTNING_INTERNAL_CONVERSATION", "false").lower() == "true"
+        body = {
+            "message": {
+                "author": {"role": "user"},
+                "content": [
+                    {"contentType": "text", "parts": [prompt]},
+                ],
+            },
+            "max_completion_tokens": max_completion_tokens,
+            "conversation_id": conversation_id,
+            "billing_project_id": billing_project_id,
+            "name": name,
+            "stream": stream,
+            "metadata": metadata or {},
+            "internal_conversation": is_internal_conversation,
+        }
+        if images:
+            for image in images:
+                url = image
+                if not image.startswith("http"):
+                    url = self._encode_image_bytes_to_data_url(image)
+
+                body["message"]["content"].append(
+                    {
+                        "contentType": "image",
+                        "parts": [url],
+                    }
+                )
+
+        if not stream:
+            thread = await asyncio.to_thread(
+                self._client.assistants_service_start_conversation, body, assistant_id, async_req=True
+            )
+            result = await asyncio.to_thread(thread.get)
+            return result.result
+
+        conversation_thread = await asyncio.to_thread(
+            self._client.assistants_service_start_conversation,
+            body,
+            assistant_id,
+            async_req=True,
+            _preload_content=False,
+        )
+
+        return self.stream_response(conversation_thread)
+
+    async def stream_response(self, thread: Any) -> AsyncGenerator[V1ConversationResponseChunk, None]:
+        loop = asyncio.get_event_loop()
+        response = await asyncio.to_thread(thread.get)
+
+        queue = asyncio.Queue()
+
+        def enqueue() -> None:
+            try:
+                for line in response:
+                    decoded_lines = line.decode("utf-8").strip()
+                    for decoded_line in decoded_lines.splitlines():
+                        chunk = self._parse_stream_line(decoded_line)
+                        if chunk:
+                            asyncio.run_coroutine_threadsafe(queue.put(chunk), loop)
+            finally:
+                asyncio.run_coroutine_threadsafe(queue.put(None), loop)
+
+        threading.Thread(target=enqueue, daemon=True).start()
+
+        while True:
+            item = await queue.get()
+            if item is None:
+                break
+            yield item
+
     def list_conversations(self, assistant_id: str) -> List[str]:
         result = self._client.assistants_service_list_conversations(assistant_id)
         return result.conversations

lightning_sdk/api/studio_api.py

@@ -154,6 +154,11 @@ class StudioApi:
     @backoff.on_exception(backoff.expo, AttributeError, max_tries=10)
     def _check_code_status_top_up_restore_finished(self, studio_id: str, teamspace_id: str) -> bool:
         """Retries checking the top_up_restore_finished value of the code status when there's an AttributeError."""
+        if (
+            self.get_studio_status(studio_id, teamspace_id) is None
+            or self.get_studio_status(studio_id, teamspace_id).in_use is None
+        ):
+            return False
         startup_status = self.get_studio_status(studio_id, teamspace_id).in_use.startup_status
         return startup_status and startup_status.top_up_restore_finished
 

lightning_sdk/cli/configure.py

@@ -17,33 +17,10 @@ def configure() -> None:
     """Configure access to resources on the Lightning AI platform."""
 
 
-@configure.command(name="ssh")
-@click.option(
-    "--name",
-    default=None,
-    help=(
-        "The name of the studio to obtain SSH config. "
-        "If not specified, tries to infer from the environment (e.g. when run from within a Studio.)"
-    ),
-)
-@click.option(
-    "--teamspace",
-    default=None,
-    help=(
-        "The teamspace the studio is part of. "
-        "Should be of format <OWNER>/<TEAMSPACE_NAME>. "
-        "If not specified, tries to infer from the environment (e.g. when run from within a Studio.)"
-    ),
-)
-@click.option(
-    "--overwrite",
-    is_flag=True,
-    flag_value=True,
-    default=False,
-    help="Whether to overwrite the SSH key and config if they already exist.",
-)
-def ssh(name: Optional[str] = None, teamspace: Optional[str] = None, overwrite: bool = False) -> None:
-    """Get SSH config entry for a studio."""
+def _configure_ssh_internal(
+    name: Optional[str] = None, teamspace: Optional[str] = None, overwrite: bool = False
+) -> None:
+    """Internal function to configure SSH without Click decorators."""
     auth = Auth()
     auth.authenticate()
     console = Console()
@@ -78,6 +55,36 @@ def ssh(name: Optional[str] = None, teamspace: Optional[str] = None, overwrite:
         console.print(f"SSH config updated at {config_path}")
 
 
+@configure.command(name="ssh")
+@click.option(
+    "--name",
+    default=None,
+    help=(
+        "The name of the studio to obtain SSH config. "
+        "If not specified, tries to infer from the environment (e.g. when run from within a Studio.)"
+    ),
+)
+@click.option(
+    "--teamspace",
+    default=None,
+    help=(
+        "The teamspace the studio is part of. "
+        "Should be of format <OWNER>/<TEAMSPACE_NAME>. "
+        "If not specified, tries to infer from the environment (e.g. when run from within a Studio.)"
+    ),
+)
+@click.option(
+    "--overwrite",
+    is_flag=True,
+    flag_value=True,
+    default=False,
+    help="Whether to overwrite the SSH key and config if they already exist.",
+)
+def ssh(name: Optional[str] = None, teamspace: Optional[str] = None, overwrite: bool = False) -> None:
+    """Get SSH config entry for a studio."""
+    _configure_ssh_internal(name=name, teamspace=teamspace, overwrite=overwrite)
+
+
 def _download_file(url: str, local_path: Path, overwrite: bool = True, chmod: Optional[int] = None) -> None:
     """Download a file from a URL."""
     import requests

lightning_sdk/cli/connect.py

@@ -4,7 +4,7 @@ from typing import Optional
 
 import click
 
-from lightning_sdk.cli.configure import ssh as configure_ssh
+from lightning_sdk.cli.configure import _configure_ssh_internal
 from lightning_sdk.cli.studios_menu import _StudiosMenu
 
 
@@ -22,7 +22,7 @@ def connect() -> None:
 )
 def studio(name: Optional[str], teamspace: Optional[str]) -> None:
     """Connect to a studio via SSH."""
-    configure_ssh(name=name, teamspace=teamspace, overwrite=False)
+    _configure_ssh_internal(name=name, teamspace=teamspace, overwrite=False)
 
     menu = _StudiosMenu()
     studio = menu._get_studio(name=name, teamspace=teamspace)

lightning_sdk/deployment/deployment.py

@@ -276,8 +276,8 @@ class Deployment:
             cloud_account=cloud_account,
             machine=machine,
             image=image,
-            entrypoint=entrypoint or "",
-            command=command or "",
+            entrypoint=entrypoint,
+            command=command,
             ports=ports,
             custom_domain=custom_domain,
             auth=auth,
@@ -340,7 +340,7 @@ class Deployment:
         return None
 
     @property
-    def readiness_probe(self) -> Optional[Union[HttpHealthCheck, ExecHealthCheck]]:
+    def health_check(self) -> Optional[Union[HttpHealthCheck, ExecHealthCheck]]:
         """The health check to validate the replicas are ready to receive traffic."""
         if self._deployment:
             self._deployment = self._deployment_api.get_deployment_by_name(self._name, self._teamspace.id)
@@ -467,6 +467,20 @@ class Deployment:
             return self._deployment.spec.image
         return None
 
+    @property
+    def entrypoint(self) -> Optional[str]:
+        if self._deployment:
+            self._deployment = self._deployment_api.get_deployment_by_name(self._name, self._teamspace.id)
+            return self._deployment.spec.entrypoint
+        return None
+
+    @property
+    def command(self) -> Optional[str]:
+        if self._deployment:
+            self._deployment = self._deployment_api.get_deployment_by_name(self._name, self._teamspace.id)
+            return self._deployment.spec.command
+        return None
+
     @property
     def _session(self) -> Any:
         if self._request_session is None:

lightning_sdk/lightning_cloud/openapi/__init__.py

@@ -563,7 +563,6 @@ from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_balance_response i
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_notification_preferences_response import V1GetUserNotificationPreferencesResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_response import V1GetUserResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_storage_breakdown_response import V1GetUserStorageBreakdownResponse
-from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_storage_response import V1GetUserStorageResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_git_credentials import V1GitCredentials
 from lightning_sdk.lightning_cloud.openapi.models.v1_google_cloud_direct_v1 import V1GoogleCloudDirectV1
 from lightning_sdk.lightning_cloud.openapi.models.v1_google_cloud_direct_v1_status import V1GoogleCloudDirectV1Status

lightning_sdk/lightning_cloud/openapi/api/cluster_service_api.py

@@ -2726,7 +2726,6 @@ class ClusterServiceApi(object):
         >>> result = thread.get()
 
         :param async_req bool
-        :param bool include_pricing:
         :param str cloud_provider:
         :param str project_id:
         :return: V1ListDefaultClusterAcceleratorsResponse
@@ -2749,7 +2748,6 @@ class ClusterServiceApi(object):
         >>> result = thread.get()
 
         :param async_req bool
-        :param bool include_pricing:
         :param str cloud_provider:
         :param str project_id:
         :return: V1ListDefaultClusterAcceleratorsResponse
@@ -2757,7 +2755,7 @@ class ClusterServiceApi(object):
                  returns the request thread.
         """
 
-        all_params = ['include_pricing', 'cloud_provider', 'project_id']  # noqa: E501
+        all_params = ['cloud_provider', 'project_id']  # noqa: E501
         all_params.append('async_req')
         all_params.append('_return_http_data_only')
         all_params.append('_preload_content')
@@ -2778,8 +2776,6 @@ class ClusterServiceApi(object):
         path_params = {}
 
         query_params = []
-        if 'include_pricing' in params:
-            query_params.append(('includePricing', params['include_pricing']))  # noqa: E501
         if 'cloud_provider' in params:
             query_params.append(('cloudProvider', params['cloud_provider']))  # noqa: E501
         if 'project_id' in params:

lightning_sdk/lightning_cloud/openapi/api/endpoint_service_api.py

@@ -1420,6 +1420,8 @@ class EndpointServiceApi(object):
         :param list[str] ids:
         :param bool active_cloudspaces:
         :param bool active_jobs:
+        :param list[str] cloudspace_ids:
+        :param list[str] job_ids:
         :return: V1ListEndpointsResponse
                  If the method is called asynchronously,
                  returns the request thread.
@@ -1447,12 +1449,14 @@ class EndpointServiceApi(object):
         :param list[str] ids:
         :param bool active_cloudspaces:
         :param bool active_jobs:
+        :param list[str] cloudspace_ids:
+        :param list[str] job_ids:
         :return: V1ListEndpointsResponse
                  If the method is called asynchronously,
                  returns the request thread.
         """
 
-        all_params = ['project_id', 'cloudspace_id', 'auto_start', 'cluster_id', 'ids', 'active_cloudspaces', 'active_jobs']  # noqa: E501
+        all_params = ['project_id', 'cloudspace_id', 'auto_start', 'cluster_id', 'ids', 'active_cloudspaces', 'active_jobs', 'cloudspace_ids', 'job_ids']  # noqa: E501
         all_params.append('async_req')
         all_params.append('_return_http_data_only')
         all_params.append('_preload_content')
@@ -1492,6 +1496,12 @@ class EndpointServiceApi(object):
             query_params.append(('activeCloudspaces', params['active_cloudspaces']))  # noqa: E501
         if 'active_jobs' in params:
             query_params.append(('activeJobs', params['active_jobs']))  # noqa: E501
+        if 'cloudspace_ids' in params:
+            query_params.append(('cloudspaceIds', params['cloudspace_ids']))  # noqa: E501
+            collection_formats['cloudspaceIds'] = 'multi'  # noqa: E501
+        if 'job_ids' in params:
+            query_params.append(('jobIds', params['job_ids']))  # noqa: E501
+            collection_formats['jobIds'] = 'multi'  # noqa: E501
 
         header_params = {}
 

lightning_sdk/lightning_cloud/openapi/api/user_service_api.py

@@ -706,91 +706,6 @@ class UserServiceApi(object):
             _request_timeout=params.get('_request_timeout'),
             collection_formats=collection_formats)
 
-    def user_service_get_user_storage(self, **kwargs) -> 'V1GetUserStorageResponse':  # noqa: E501
-        """user_service_get_user_storage  # noqa: E501
-
-        This method makes a synchronous HTTP request by default. To make an
-        asynchronous HTTP request, please pass async_req=True
-        >>> thread = api.user_service_get_user_storage(async_req=True)
-        >>> result = thread.get()
-
-        :param async_req bool
-        :return: V1GetUserStorageResponse
-                 If the method is called asynchronously,
-                 returns the request thread.
-        """
-        kwargs['_return_http_data_only'] = True
-        if kwargs.get('async_req'):
-            return self.user_service_get_user_storage_with_http_info(**kwargs)  # noqa: E501
-        else:
-            (data) = self.user_service_get_user_storage_with_http_info(**kwargs)  # noqa: E501
-            return data
-
-    def user_service_get_user_storage_with_http_info(self, **kwargs) -> 'V1GetUserStorageResponse':  # noqa: E501
-        """user_service_get_user_storage  # noqa: E501
-
-        This method makes a synchronous HTTP request by default. To make an
-        asynchronous HTTP request, please pass async_req=True
-        >>> thread = api.user_service_get_user_storage_with_http_info(async_req=True)
-        >>> result = thread.get()
-
-        :param async_req bool
-        :return: V1GetUserStorageResponse
-                 If the method is called asynchronously,
-                 returns the request thread.
-        """
-
-        all_params = []  # noqa: E501
-        all_params.append('async_req')
-        all_params.append('_return_http_data_only')
-        all_params.append('_preload_content')
-        all_params.append('_request_timeout')
-
-        params = locals()
-        for key, val in six.iteritems(params['kwargs']):
-            if key not in all_params:
-                raise TypeError(
-                    "Got an unexpected keyword argument '%s'"
-                    " to method user_service_get_user_storage" % key
-                )
-            params[key] = val
-        del params['kwargs']
-
-        collection_formats = {}
-
-        path_params = {}
-
-        query_params = []
-
-        header_params = {}
-
-        form_params = []
-        local_var_files = {}
-
-        body_params = None
-        # HTTP header `Accept`
-        header_params['Accept'] = self.api_client.select_header_accept(
-            ['application/json'])  # noqa: E501
-
-        # Authentication setting
-        auth_settings = []  # noqa: E501
-
-        return self.api_client.call_api(
-            '/v1/users/storage', 'GET',
-            path_params,
-            query_params,
-            header_params,
-            body=body_params,
-            post_params=form_params,
-            files=local_var_files,
-            response_type='V1GetUserStorageResponse',  # noqa: E501
-            auth_settings=auth_settings,
-            async_req=params.get('async_req'),
-            _return_http_data_only=params.get('_return_http_data_only'),
-            _preload_content=params.get('_preload_content', True),
-            _request_timeout=params.get('_request_timeout'),
-            collection_formats=collection_formats)
-
     def user_service_get_user_storage_breakdown(self, **kwargs) -> 'V1GetUserStorageBreakdownResponse':  # noqa: E501
         """user_service_get_user_storage_breakdown  # noqa: E501
 

lightning_sdk/lightning_cloud/openapi/models/__init__.py

@@ -518,7 +518,6 @@ from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_balance_response i
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_notification_preferences_response import V1GetUserNotificationPreferencesResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_response import V1GetUserResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_storage_breakdown_response import V1GetUserStorageBreakdownResponse
-from lightning_sdk.lightning_cloud.openapi.models.v1_get_user_storage_response import V1GetUserStorageResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_git_credentials import V1GitCredentials
 from lightning_sdk.lightning_cloud.openapi.models.v1_google_cloud_direct_v1 import V1GoogleCloudDirectV1
 from lightning_sdk.lightning_cloud.openapi.models.v1_google_cloud_direct_v1_status import V1GoogleCloudDirectV1Status