together 2.0.0a16__py3-none-any.whl → 2.0.0a18__py3-none-any.whl

together/_base_client.py

@@ -86,6 +86,7 @@ from ._exceptions import (
     APIConnectionError,
     APIResponseValidationError,
 )
+from ._utils._json import openapi_dumps
 
 log: logging.Logger = logging.getLogger(__name__)
 
@@ -554,8 +555,10 @@ class BaseClient(Generic[_HttpxClientT, _DefaultStreamT]):
                 kwargs["content"] = options.content
             elif isinstance(json_data, bytes):
                 kwargs["content"] = json_data
-            else:
-                kwargs["json"] = json_data if is_given(json_data) else None
+            elif not files:
+                # Don't set content when JSON is sent as multipart/form-data,
+                # since httpx's content param overrides other body arguments
+                kwargs["content"] = openapi_dumps(json_data) if is_given(json_data) and json_data is not None else None
             kwargs["files"] = files
         else:
             headers.pop("Content-Type", None)

together/_client.py

@@ -37,7 +37,6 @@ if TYPE_CHECKING:
     from .resources import (
         beta,
         chat,
-        jobs,
         audio,
         evals,
         files,
@@ -53,11 +52,9 @@ if TYPE_CHECKING:
         fine_tuning,
         code_interpreter,
     )
-    from .resources.jobs import JobsResource, AsyncJobsResource
     from .resources.evals import EvalsResource, AsyncEvalsResource
     from .resources.files import FilesResource, AsyncFilesResource
     from .resources.images import ImagesResource, AsyncImagesResource
-    from .resources.models import ModelsResource, AsyncModelsResource
     from .resources.rerank import RerankResource, AsyncRerankResource
     from .resources.videos import VideosResource, AsyncVideosResource
     from .resources.batches import BatchesResource, AsyncBatchesResource
@@ -69,6 +66,7 @@ if TYPE_CHECKING:
     from .resources.audio.audio import AudioResource, AsyncAudioResource
     from .resources.completions import CompletionsResource, AsyncCompletionsResource
     from .resources.fine_tuning import FineTuningResource, AsyncFineTuningResource
+    from .resources.models.models import ModelsResource, AsyncModelsResource
     from .resources.code_interpreter.code_interpreter import CodeInterpreterResource, AsyncCodeInterpreterResource
 
 __all__ = [
@@ -209,12 +207,6 @@ class Together(SyncAPIClient):
 
         return ModelsResource(self)
 
-    @cached_property
-    def jobs(self) -> JobsResource:
-        from .resources.jobs import JobsResource
-
-        return JobsResource(self)
-
     @cached_property
     def endpoints(self) -> EndpointsResource:
         from .resources.endpoints import EndpointsResource
@@ -486,12 +478,6 @@ class AsyncTogether(AsyncAPIClient):
 
         return AsyncModelsResource(self)
 
-    @cached_property
-    def jobs(self) -> AsyncJobsResource:
-        from .resources.jobs import AsyncJobsResource
-
-        return AsyncJobsResource(self)
-
     @cached_property
     def endpoints(self) -> AsyncEndpointsResource:
         from .resources.endpoints import AsyncEndpointsResource
@@ -709,12 +695,6 @@ class TogetherWithRawResponse:
 
         return ModelsResourceWithRawResponse(self._client.models)
 
-    @cached_property
-    def jobs(self) -> jobs.JobsResourceWithRawResponse:
-        from .resources.jobs import JobsResourceWithRawResponse
-
-        return JobsResourceWithRawResponse(self._client.jobs)
-
     @cached_property
     def endpoints(self) -> endpoints.EndpointsResourceWithRawResponse:
         from .resources.endpoints import EndpointsResourceWithRawResponse
@@ -818,12 +798,6 @@ class AsyncTogetherWithRawResponse:
 
         return AsyncModelsResourceWithRawResponse(self._client.models)
 
-    @cached_property
-    def jobs(self) -> jobs.AsyncJobsResourceWithRawResponse:
-        from .resources.jobs import AsyncJobsResourceWithRawResponse
-
-        return AsyncJobsResourceWithRawResponse(self._client.jobs)
-
     @cached_property
     def endpoints(self) -> endpoints.AsyncEndpointsResourceWithRawResponse:
         from .resources.endpoints import AsyncEndpointsResourceWithRawResponse
@@ -927,12 +901,6 @@ class TogetherWithStreamedResponse:
 
         return ModelsResourceWithStreamingResponse(self._client.models)
 
-    @cached_property
-    def jobs(self) -> jobs.JobsResourceWithStreamingResponse:
-        from .resources.jobs import JobsResourceWithStreamingResponse
-
-        return JobsResourceWithStreamingResponse(self._client.jobs)
-
     @cached_property
     def endpoints(self) -> endpoints.EndpointsResourceWithStreamingResponse:
         from .resources.endpoints import EndpointsResourceWithStreamingResponse
@@ -1036,12 +1004,6 @@ class AsyncTogetherWithStreamedResponse:
 
         return AsyncModelsResourceWithStreamingResponse(self._client.models)
 
-    @cached_property
-    def jobs(self) -> jobs.AsyncJobsResourceWithStreamingResponse:
-        from .resources.jobs import AsyncJobsResourceWithStreamingResponse
-
-        return AsyncJobsResourceWithStreamingResponse(self._client.jobs)
-
     @cached_property
     def endpoints(self) -> endpoints.AsyncEndpointsResourceWithStreamingResponse:
         from .resources.endpoints import AsyncEndpointsResourceWithStreamingResponse

together/_compat.py

@@ -139,6 +139,7 @@ def model_dump(
     exclude_defaults: bool = False,
     warnings: bool = True,
     mode: Literal["json", "python"] = "python",
+    by_alias: bool | None = None,
 ) -> dict[str, Any]:
     if (not PYDANTIC_V1) or hasattr(model, "model_dump"):
         return model.model_dump(
@@ -148,13 +149,12 @@ def model_dump(
             exclude_defaults=exclude_defaults,
             # warnings are not supported in Pydantic v1
             warnings=True if PYDANTIC_V1 else warnings,
+            by_alias=by_alias,
         )
     return cast(
         "dict[str, Any]",
         model.dict(  # pyright: ignore[reportDeprecated, reportUnnecessaryCast]
-            exclude=exclude,
-            exclude_unset=exclude_unset,
-            exclude_defaults=exclude_defaults,
+            exclude=exclude, exclude_unset=exclude_unset, exclude_defaults=exclude_defaults, by_alias=bool(by_alias)
         ),
     )
 

together/_utils/_json.py

@@ -0,0 +1,35 @@
+import json
+from typing import Any
+from datetime import datetime
+from typing_extensions import override
+
+import pydantic
+
+from .._compat import model_dump
+
+
+def openapi_dumps(obj: Any) -> bytes:
+    """
+    Serialize an object to UTF-8 encoded JSON bytes.
+
+    Extends the standard json.dumps with support for additional types
+    commonly used in the SDK, such as `datetime`, `pydantic.BaseModel`, etc.
+    """
+    return json.dumps(
+        obj,
+        cls=_CustomEncoder,
+        # Uses the same defaults as httpx's JSON serialization
+        ensure_ascii=False,
+        separators=(",", ":"),
+        allow_nan=False,
+    ).encode()
+
+
+class _CustomEncoder(json.JSONEncoder):
+    @override
+    def default(self, o: Any) -> Any:
+        if isinstance(o, datetime):
+            return o.isoformat()
+        if isinstance(o, pydantic.BaseModel):
+            return model_dump(o, exclude_unset=True, mode="json", by_alias=True)
+        return super().default(o)

together/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "together"
-__version__ = "2.0.0-alpha.16"  # x-release-please-version
+__version__ = "2.0.0-alpha.18"  # x-release-please-version

together/lib/cli/api/endpoints/create.py

@@ -3,7 +3,6 @@ from __future__ import annotations
 import sys
 
 import click
-from rich import print
 
 from together import APIError, Together, omit
 from together.lib.cli.api._utils import handle_api_errors
@@ -31,7 +30,7 @@ from .hardware import hardware
 )
 @click.option(
     "--gpu",
-    type=click.Choice(["h100", "a100", "l40", "l40s", "rtx-6000"]),
+    type=click.Choice(["b200", "h200", "h100", "a100", "l40", "l40s", "rtx-6000"]),
     required=True,
     help="GPU type to use for inference",
 )
@@ -48,7 +47,7 @@ from .hardware import hardware
 @click.option(
     "--no-prompt-cache",
     is_flag=True,
-    help="Disable the prompt cache for this endpoint",
+    help="Deprecated and no longer has any effect.",
 )
 @click.option(
     "--no-speculative-decoding",
@@ -95,6 +94,8 @@ def create(
     client: Together = ctx.obj
     # Map GPU types to their full hardware ID names
     gpu_map = {
+        "b200": "nvidia_b200_180gb_sxm",
+        "h200": "nvidia_h200_140gb_sxm",
         "h100": "nvidia_h100_80gb_sxm",
         "a100": "nvidia_a100_80gb_pcie" if gpu_count == 1 else "nvidia_a100_80gb_sxm",
         "l40": "nvidia_l40",
@@ -102,6 +103,9 @@ def create(
         "rtx-6000": "nvidia_rtx_6000_ada",
     }
 
+    if no_prompt_cache is not None:
+        click.echo("Warning: --no-prompt-cache is deprecated and no longer has any effect.", err=True)
+
     hardware_id = f"{gpu_count}x_{gpu_map[gpu]}"
 
     try:
@@ -113,15 +117,19 @@ def create(
                 "max_replicas": max_replicas,
             },
             display_name=display_name or omit,
-            disable_prompt_cache=no_prompt_cache or omit,
             disable_speculative_decoding=no_speculative_decoding or omit,
             state="STOPPED" if no_auto_start else "STARTED",
             inactive_timeout=inactive_timeout,
             extra_query={"availability_zone": availability_zone or omit},
         )
     except APIError as e:
-        if "check the hardware api" in str(e.args[0]).lower() or "invalid hardware provided" in str(e.args[0]).lower():
-            print("Invalid hardware provided")
+        if (
+            "check the hardware api" in str(e.args[0]).lower()
+            or "invalid hardware provided" in str(e.args[0]).lower()
+            or "the selected configuration" in str(e.args[0]).lower()
+        ):
+            click.secho("Invalid hardware selected.", fg="red", err=True)
+            click.echo("\nAvailable hardware options:")
             ctx.invoke(hardware, available=True, model=model, json=False)
             sys.exit(1)
         raise e
@@ -134,8 +142,6 @@ def create(
     click.echo(f"  Hardware: {hardware_id}", err=True)
     if display_name:
         click.echo(f"  Display name: {display_name}", err=True)
-    if no_prompt_cache:
-        click.echo("  Prompt cache: disabled", err=True)
     if no_speculative_decoding:
         click.echo("  Speculative decoding: disabled", err=True)
     if no_auto_start:

together/lib/cli/api/endpoints/hardware.py

@@ -1,10 +1,14 @@
 from __future__ import annotations
 
+import re
 import json as json_lib
+from typing import Any, Dict, List
 
 import click
+from tabulate import tabulate
 
 from together import Together, omit
+from together.types import HardwareListResponse
 from together.lib.cli.api._utils import handle_api_errors
 from together.lib.utils.serializer import datetime_serializer
 
@@ -21,20 +25,47 @@ from together.lib.utils.serializer import datetime_serializer
 @handle_api_errors("Endpoints")
 def hardware(client: Together, model: str | None, json: bool, available: bool) -> None:
     """List all available hardware options, optionally filtered by model."""
-    message = "Available hardware options:" if available else "All hardware options:"
-    click.echo(message, err=True)
     hardware_options = client.hardware.list(model=model or omit)
-    # hardware_options = client.endpoints.list_hardware(model)
+
     if available:
         hardware_options.data = [
             hardware
             for hardware in hardware_options.data
             if hardware.availability is not None and hardware.availability.status == "available"
         ]
-
     if json:
         json_output = [hardware.model_dump() for hardware in hardware_options.data]
         click.echo(json_lib.dumps(json_output, default=datetime_serializer, indent=2))
     else:
-        for hardware in hardware_options.data:
-            click.echo(f"  {hardware.id}", err=True)
+        _format_hardware_options(hardware_options, show_availability=model is not None)
+
+
+def _format_hardware_options(hardware_options: HardwareListResponse, show_availability: bool = True) -> None:
+    display_list: List[Dict[str, Any]] = []
+
+    for hw in hardware_options.data:
+        data = {
+            "Hardware ID": hw.id,
+            "GPU": re.sub(r"\-\d+[a-zA-Z][a-zA-Z]$", "", hw.specs.gpu_type)
+            if hw.specs and hw.specs.gpu_type
+            else "N/A",
+            "Memory": f"{int(hw.specs.gpu_memory)}GB" if hw.specs else "N/A",
+            "Count": hw.specs.gpu_count if hw.specs else "N/A",
+            "Price (per minute)": (f"${hw.pricing.cents_per_minute / 100:.2f}" if hw.pricing else "N/A"),
+        }
+
+        if show_availability:
+            status_display = "—"
+            if hw.availability:
+                status = hw.availability.status
+                # Add visual indicators for status
+                if status == "available":
+                    status_display = click.style("✓ available", fg="green")
+                elif status == "unavailable":
+                    status_display = click.style("✗ unavailable", fg="red")
+                else:  # insufficient
+                    status_display = click.style("⚠ insufficient", fg="yellow")
+                data["availability"] = status_display
+        display_list.append(data)
+
+    click.echo(tabulate(display_list, headers="keys", numalign="left"))

together/lib/cli/api/models/list.py

@@ -5,9 +5,9 @@ import click
 from tabulate import tabulate
 
 from together import Together, omit
-from together._models import BaseModel
 from together._response import APIResponse as APIResponse
 from together.lib.cli.api._utils import handle_api_errors
+from together.lib.utils.serializer import datetime_serializer
 
 
 @click.command()
@@ -29,23 +29,27 @@ def list(ctx: click.Context, type: Optional[str], json: bool) -> None:
 
     models_list = client.models.list(dedicated=type == "dedicated" if type else omit)
 
+    if json:
+        items = [model.model_dump() for model in models_list]
+        click.echo(json_lib.dumps(items, indent=2, default=datetime_serializer))
+        return
+
     display_list: List[Dict[str, Any]] = []
-    model: BaseModel
-    for model in models_list:
+    for model in sorted(models_list, key=lambda x: x.type):
+        price_parts: List[str] = []
+
+        # Only show pricing if a value actually exists
+        if model.pricing and model.pricing.input > 0 and model.pricing.output > 0:
+            price_parts.append(f"${model.pricing.input:.2f}")
+            price_parts.append(f"${model.pricing.output:.2f}")
+
         display_list.append(
             {
-                "ID": model.id,
-                "Name": model.display_name,
-                "Organization": model.organization,
+                "Model": model.id,
                 "Type": model.type,
-                "Context Length": model.context_length,
-                "License": model.license,
-                "Input per 1M token": model.pricing.input if model.pricing else None,
-                "Output per 1M token": model.pricing.output if model.pricing else None,
+                "Context length": model.context_length if model.context_length else None,
+                "Price per 1M Tokens (input/output)": "/".join(price_parts),
             }
         )
 
-    if json:
-        click.echo(json_lib.dumps(display_list, indent=2))
-    else:
-        click.echo(tabulate(display_list, headers="keys", tablefmt="plain"))
+    click.echo(tabulate(display_list, headers="keys"))

together/lib/cli/api/models/upload.py

@@ -3,7 +3,7 @@ from typing import Literal, Optional
 
 import click
 
-from together import Together, omit
+from together import Together, TogetherError, omit
 from together._response import APIResponse as APIResponse
 from together.lib.cli.api._utils import handle_api_errors
 from together.types.model_upload_response import ModelUploadResponse
@@ -76,6 +76,10 @@ def upload(
     if json:
         click.echo(json_lib.dumps(response.model_dump(), indent=2))
     else:
+        # If the model weights already exist, the api is returning 200 but with no data
+        if response.data is None: # type: ignore
+            raise TogetherError(response.message)
+
         click.echo(f"Model upload job created successfully!")
         if response.data.job_id:
             click.echo(f"Job ID: {response.data.job_id}")

together/resources/__init__.py

@@ -16,14 +16,6 @@ from .chat import (
     ChatResourceWithStreamingResponse,
     AsyncChatResourceWithStreamingResponse,
 )
-from .jobs import (
-    JobsResource,
-    AsyncJobsResource,
-    JobsResourceWithRawResponse,
-    AsyncJobsResourceWithRawResponse,
-    JobsResourceWithStreamingResponse,
-    AsyncJobsResourceWithStreamingResponse,
-)
 from .audio import (
     AudioResource,
     AsyncAudioResource,
@@ -204,12 +196,6 @@ __all__ = [
     "AsyncModelsResourceWithRawResponse",
     "ModelsResourceWithStreamingResponse",
     "AsyncModelsResourceWithStreamingResponse",
-    "JobsResource",
-    "AsyncJobsResource",
-    "JobsResourceWithRawResponse",
-    "AsyncJobsResourceWithRawResponse",
-    "JobsResourceWithStreamingResponse",
-    "AsyncJobsResourceWithStreamingResponse",
     "EndpointsResource",
     "AsyncEndpointsResource",
     "EndpointsResourceWithRawResponse",

together/resources/beta/__init__.py

@@ -1,5 +1,13 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
+from .jig import (
+    JigResource,
+    AsyncJigResource,
+    JigResourceWithRawResponse,
+    AsyncJigResourceWithRawResponse,
+    JigResourceWithStreamingResponse,
+    AsyncJigResourceWithStreamingResponse,
+)
 from .beta import (
     BetaResource,
     AsyncBetaResource,
@@ -18,6 +26,12 @@ from .clusters import (
 )
 
 __all__ = [
+    "JigResource",
+    "AsyncJigResource",
+    "JigResourceWithRawResponse",
+    "AsyncJigResourceWithRawResponse",
+    "JigResourceWithStreamingResponse",
+    "AsyncJigResourceWithStreamingResponse",
     "ClustersResource",
     "AsyncClustersResource",
     "ClustersResourceWithRawResponse",

together/resources/beta/beta.py

@@ -2,6 +2,14 @@
 
 from __future__ import annotations
 
+from .jig.jig import (
+    JigResource,
+    AsyncJigResource,
+    JigResourceWithRawResponse,
+    AsyncJigResourceWithRawResponse,
+    JigResourceWithStreamingResponse,
+    AsyncJigResourceWithStreamingResponse,
+)
 from ..._compat import cached_property
 from ..._resource import SyncAPIResource, AsyncAPIResource
 from .clusters.clusters import (
@@ -17,6 +25,10 @@ __all__ = ["BetaResource", "AsyncBetaResource"]
 
 
 class BetaResource(SyncAPIResource):
+    @cached_property
+    def jig(self) -> JigResource:
+        return JigResource(self._client)
+
     @cached_property
     def clusters(self) -> ClustersResource:
         return ClustersResource(self._client)
@@ -42,6 +54,10 @@ class BetaResource(SyncAPIResource):
 
 
 class AsyncBetaResource(AsyncAPIResource):
+    @cached_property
+    def jig(self) -> AsyncJigResource:
+        return AsyncJigResource(self._client)
+
     @cached_property
     def clusters(self) -> AsyncClustersResource:
         return AsyncClustersResource(self._client)
@@ -70,6 +86,10 @@ class BetaResourceWithRawResponse:
     def __init__(self, beta: BetaResource) -> None:
         self._beta = beta
 
+    @cached_property
+    def jig(self) -> JigResourceWithRawResponse:
+        return JigResourceWithRawResponse(self._beta.jig)
+
     @cached_property
     def clusters(self) -> ClustersResourceWithRawResponse:
         return ClustersResourceWithRawResponse(self._beta.clusters)
@@ -79,6 +99,10 @@ class AsyncBetaResourceWithRawResponse:
     def __init__(self, beta: AsyncBetaResource) -> None:
         self._beta = beta
 
+    @cached_property
+    def jig(self) -> AsyncJigResourceWithRawResponse:
+        return AsyncJigResourceWithRawResponse(self._beta.jig)
+
     @cached_property
     def clusters(self) -> AsyncClustersResourceWithRawResponse:
         return AsyncClustersResourceWithRawResponse(self._beta.clusters)
@@ -88,6 +112,10 @@ class BetaResourceWithStreamingResponse:
     def __init__(self, beta: BetaResource) -> None:
         self._beta = beta
 
+    @cached_property
+    def jig(self) -> JigResourceWithStreamingResponse:
+        return JigResourceWithStreamingResponse(self._beta.jig)
+
     @cached_property
     def clusters(self) -> ClustersResourceWithStreamingResponse:
         return ClustersResourceWithStreamingResponse(self._beta.clusters)
@@ -97,6 +125,10 @@ class AsyncBetaResourceWithStreamingResponse:
     def __init__(self, beta: AsyncBetaResource) -> None:
         self._beta = beta
 
+    @cached_property
+    def jig(self) -> AsyncJigResourceWithStreamingResponse:
+        return AsyncJigResourceWithStreamingResponse(self._beta.jig)
+
     @cached_property
     def clusters(self) -> AsyncClustersResourceWithStreamingResponse:
         return AsyncClustersResourceWithStreamingResponse(self._beta.clusters)

together/resources/beta/clusters/clusters.py

@@ -110,7 +110,7 @@ class ClustersResource(SyncAPIResource):
           timeout: Override the client-level default timeout for this request, in seconds
         """
         return self._post(
-            "/clusters",
+            "/compute/clusters",
             body=maybe_transform(
                 {
                     "billing_type": billing_type,
@@ -158,7 +158,7 @@ class ClustersResource(SyncAPIResource):
         if not cluster_id:
             raise ValueError(f"Expected a non-empty value for `cluster_id` but received {cluster_id!r}")
         return self._get(
-            f"/clusters/{cluster_id}",
+            f"/compute/clusters/{cluster_id}",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -193,7 +193,7 @@ class ClustersResource(SyncAPIResource):
         if not cluster_id:
             raise ValueError(f"Expected a non-empty value for `cluster_id` but received {cluster_id!r}")
         return self._put(
-            f"/clusters/{cluster_id}",
+            f"/compute/clusters/{cluster_id}",
             body=maybe_transform(
                 {
                     "cluster_type": cluster_type,
@@ -219,7 +219,7 @@ class ClustersResource(SyncAPIResource):
     ) -> ClusterListResponse:
         """List all GPU clusters."""
         return self._get(
-            "/clusters",
+            "/compute/clusters",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -252,7 +252,7 @@ class ClustersResource(SyncAPIResource):
         if not cluster_id:
             raise ValueError(f"Expected a non-empty value for `cluster_id` but received {cluster_id!r}")
         return self._delete(
-            f"/clusters/{cluster_id}",
+            f"/compute/clusters/{cluster_id}",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -271,7 +271,7 @@ class ClustersResource(SyncAPIResource):
     ) -> ClusterListRegionsResponse:
         """List regions and corresponding supported driver versions"""
         return self._get(
-            "/clusters/regions",
+            "/compute/regions",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -355,7 +355,7 @@ class AsyncClustersResource(AsyncAPIResource):
           timeout: Override the client-level default timeout for this request, in seconds
         """
         return await self._post(
-            "/clusters",
+            "/compute/clusters",
             body=await async_maybe_transform(
                 {
                     "billing_type": billing_type,
@@ -403,7 +403,7 @@ class AsyncClustersResource(AsyncAPIResource):
         if not cluster_id:
             raise ValueError(f"Expected a non-empty value for `cluster_id` but received {cluster_id!r}")
         return await self._get(
-            f"/clusters/{cluster_id}",
+            f"/compute/clusters/{cluster_id}",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -438,7 +438,7 @@ class AsyncClustersResource(AsyncAPIResource):
         if not cluster_id:
             raise ValueError(f"Expected a non-empty value for `cluster_id` but received {cluster_id!r}")
         return await self._put(
-            f"/clusters/{cluster_id}",
+            f"/compute/clusters/{cluster_id}",
             body=await async_maybe_transform(
                 {
                     "cluster_type": cluster_type,
@@ -464,7 +464,7 @@ class AsyncClustersResource(AsyncAPIResource):
     ) -> ClusterListResponse:
         """List all GPU clusters."""
         return await self._get(
-            "/clusters",
+            "/compute/clusters",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -497,7 +497,7 @@ class AsyncClustersResource(AsyncAPIResource):
         if not cluster_id:
             raise ValueError(f"Expected a non-empty value for `cluster_id` but received {cluster_id!r}")
         return await self._delete(
-            f"/clusters/{cluster_id}",
+            f"/compute/clusters/{cluster_id}",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
@@ -516,7 +516,7 @@ class AsyncClustersResource(AsyncAPIResource):
     ) -> ClusterListRegionsResponse:
         """List regions and corresponding supported driver versions"""
         return await self._get(
-            "/clusters/regions",
+            "/compute/regions",
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),