add list avzone and deploy de in avzone

src/together/cli/api/endpoints.py

@@ -132,6 +132,10 @@ def endpoints(ctx: click.Context) -> None:
     type=int,
     help="Number of minutes of inactivity after which the endpoint will be automatically stopped. Set to 0 to disable.",
 )
+@click.option(
+    "--user-specified-avzone",
+    help="User-specified availability zone (e.g., us-central-4b)",
+)
 @click.option(
     "--wait",
     is_flag=True,
@@ -152,6 +156,7 @@ def create(
     no_speculative_decoding: bool,
     no_auto_start: bool,
     inactive_timeout: int | None,
+    user_specified_avzone: str | None,
     wait: bool,
 ) -> None:
     """Create a new dedicated inference endpoint."""
@@ -177,6 +182,7 @@ def create(
             disable_speculative_decoding=no_speculative_decoding,
             state="STOPPED" if no_auto_start else "STARTED",
             inactive_timeout=inactive_timeout,
+            user_specified_avzone=user_specified_avzone,
         )
     except InvalidRequestError as e:
         print_api_error(e)
@@ -203,6 +209,8 @@ def create(
         click.echo("  Auto-start: disabled", err=True)
     if inactive_timeout is not None:
         click.echo(f"  Inactive timeout: {inactive_timeout} minutes", err=True)
+    if user_specified_avzone:
+        click.echo(f"  Availability zone: {user_specified_avzone}", err=True)
 
     click.echo(f"Endpoint created successfully, id: {response.id}", err=True)
 
@@ -432,3 +440,23 @@ def update(
 
     click.echo("Successfully updated endpoint", err=True)
     click.echo(endpoint_id)
+
+@endpoints.command()
+@click.option("--json", is_flag=True, help="Print output in JSON format")
+@click.pass_obj
+@handle_api_errors
+def avzones(client: Together, json: bool) -> None:
+    """List all available availability zones."""
+    avzones = client.endpoints.list_avzones()
+
+    if not avzones:
+        click.echo("No availability zones found", err=True)
+        return
+
+    if json:
+        import json as json_lib
+        click.echo(json_lib.dumps({"avzones": avzones}, indent=2))
+    else:
+        click.echo("Available zones:", err=True)
+        for avzone in sorted(avzones):
+            click.echo(f"  {avzone}")

src/together/resources/endpoints.py

@@ -60,6 +60,7 @@ def create(
         disable_speculative_decoding: bool = True,
         state: Literal["STARTED", "STOPPED"] = "STARTED",
         inactive_timeout: Optional[int] = None,
+        user_specified_avzone: Optional[str] = None,
     ) -> DedicatedEndpoint:
         """
         Create a new dedicated endpoint.
@@ -74,6 +75,7 @@ def create(
             disable_speculative_decoding (bool, optional): Whether to disable speculative decoding. Defaults to False.
             state (str, optional): The desired state of the endpoint. Defaults to "STARTED".
             inactive_timeout (int, optional): The number of minutes of inactivity after which the endpoint will be automatically stopped. Set to 0 to disable automatic timeout.
+            user_specified_avzone (str, optional): Start endpoint in specified availability zone (e.g., us-central-4b).
 
         Returns:
             DedicatedEndpoint: Object containing endpoint information
@@ -100,6 +102,9 @@ def create(
         if inactive_timeout is not None:
             data["inactive_timeout"] = inactive_timeout
 
+        if user_specified_avzone is not None:
+            data["user_specified_avzone"] = user_specified_avzone
+
         response, _, _ = requestor.request(
             options=TogetherRequest(
                 method="POST",
@@ -257,6 +262,31 @@ def list_hardware(self, model: Optional[str] = None) -> List[HardwareWithStatus]
 
         return [HardwareWithStatus(**item) for item in response.data["data"]]
 
+    def list_avzones(self) -> List[str]:
+        """
+        List all available availability zones.
+
+        Returns:
+            List[str]: List of unique availability zones
+        """
+        requestor = api_requestor.APIRequestor(
+            client=self._client,
+        )
+
+        response, _, _ = requestor.request(
+            options=TogetherRequest(
+                method="GET",
+                url="clusters/avzones",
+            ),
+            stream=False,
+        )
+
+        assert isinstance(response, TogetherResponse)
+        assert isinstance(response.data, dict)
+        assert isinstance(response.data["avzones"], list)
+
+        return response.data["avzones"]
+
 
 class AsyncEndpoints:
     def __init__(self, client: TogetherClient) -> None:
@@ -308,6 +338,7 @@ async def create(
         disable_speculative_decoding: bool = True,
         state: Literal["STARTED", "STOPPED"] = "STARTED",
         inactive_timeout: Optional[int] = None,
+        user_specified_avzone: Optional[str] = None,
     ) -> DedicatedEndpoint:
         """
         Create a new dedicated endpoint.
@@ -348,6 +379,9 @@ async def create(
         if inactive_timeout is not None:
             data["inactive_timeout"] = inactive_timeout
 
+        if user_specified_avzone is not None:
+            data["user_specified_avzone"] = user_specified_avzone
+
         response, _, _ = await requestor.arequest(
             options=TogetherRequest(
                 method="POST",
@@ -506,3 +540,28 @@ async def list_hardware(
         assert isinstance(response.data["data"], list)
 
         return [HardwareWithStatus(**item) for item in response.data["data"]]
+
+    async def list_avzones(self) -> List[str]:
+        """
+        List all available availability zones.
+
+        Returns:
+            List[str]: List of unique availability zones
+        """
+        requestor = api_requestor.APIRequestor(
+            client=self._client,
+        )
+
+        response, _, _ = await requestor.arequest(
+            options=TogetherRequest(
+                method="GET",
+                url="clusters/avzones",
+            ),
+            stream=False,
+        )
+
+        assert isinstance(response, TogetherResponse)
+        assert isinstance(response.data, dict)
+        assert isinstance(response.data["avzones"], list)
+
+        return response.data["avzones"]