Release SDK updates

.github/workflows/ci.yml

@@ -58,3 +58,43 @@ jobs:
 
       - name: Run build
         run: uv build
+
+  test-others:
+    timeout-minutes: 10
+    name: test-others
+    if: (github.event_name == 'push' || github.event.pull_request.head.repo.fork) && (github.event_name != 'push' || github.event.head_commit.message != 'codegen metadata')
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5.4.2
+        with:
+          version: '0.10.2'
+
+      - name: Install dependencies
+        run: uv sync --all-extras
+
+      - name: Run tests
+        env:
+          TOGETHER_API_KEY: ${{ secrets.TOGETHER_API_KEY }}
+        run: './scripts/test --ignore=tests/integration'
+
+  test-integration:
+    if: (github.event_name == 'push' || github.event.pull_request.head.repo.fork) && (github.event_name != 'push' || github.event.head_commit.message != 'codegen metadata')
+    timeout-minutes: 10
+    name: test-integration
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5.4.2
+        with:
+          version: '0.10.2'
+
+      - name: Install dependencies
+        run: uv sync --all-extras
+
+      - name: Run tests
+        env:
+          TOGETHER_API_KEY: ${{ secrets.TOGETHER_API_KEY }}
+        run: './scripts/test tests/integration'

.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "2.20.0"
+  ".": "2.21.0"
 }

.stats.yml

@@ -1 +1 @@
-configured_endpoints: 85
+configured_endpoints: 86

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## [2.21.0](https://github.com/togethercomputer/together-py/compare/v2.20.0...v2.21.0) (2026-06-29)
+
+
+### Features
+
+* Add early_stopping params to fine_tuning.create() ([#425](https://github.com/togethercomputer/together-py/issues/425)) ([74a793c](https://github.com/togethercomputer/together-py/commit/74a793c2415cdb82453906195d06c4e71e06296c))
+
 ## [2.20.0](https://github.com/togethercomputer/together-py/compare/v2.19.0...v2.20.0) (2026-06-26)
 
 

api.md

@@ -37,6 +37,7 @@ Types:
 from together.types.beta.jig import (
     QueueRetrieveResponse,
     QueueCancelResponse,
+    QueueClearResponse,
     QueueMetricsResponse,
     QueueSubmitResponse,
 )
@@ -46,6 +47,7 @@ Methods:
 
 - <code title="get /queue/status">client.beta.jig.queue.<a href="./src/together/resources/beta/jig/queue.py">retrieve</a>(\*\*<a href="src/together/types/beta/jig/queue_retrieve_params.py">params</a>) -> <a href="./src/together/types/beta/jig/queue_retrieve_response.py">QueueRetrieveResponse</a></code>
 - <code title="post /queue/cancel">client.beta.jig.queue.<a href="./src/together/resources/beta/jig/queue.py">cancel</a>(\*\*<a href="src/together/types/beta/jig/queue_cancel_params.py">params</a>) -> <a href="./src/together/types/beta/jig/queue_cancel_response.py">QueueCancelResponse</a></code>
+- <code title="post /queue/clear">client.beta.jig.queue.<a href="./src/together/resources/beta/jig/queue.py">clear</a>(\*\*<a href="src/together/types/beta/jig/queue_clear_params.py">params</a>) -> <a href="./src/together/types/beta/jig/queue_clear_response.py">QueueClearResponse</a></code>
 - <code title="get /queue/metrics">client.beta.jig.queue.<a href="./src/together/resources/beta/jig/queue.py">metrics</a>(\*\*<a href="src/together/types/beta/jig/queue_metrics_params.py">params</a>) -> <a href="./src/together/types/beta/jig/queue_metrics_response.py">QueueMetricsResponse</a></code>
 - <code title="post /queue/submit">client.beta.jig.queue.<a href="./src/together/resources/beta/jig/queue.py">submit</a>(\*\*<a href="src/together/types/beta/jig/queue_submit_params.py">params</a>) -> <a href="./src/together/types/beta/jig/queue_submit_response.py">QueueSubmitResponse</a></code>
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "together"
-version = "2.20.0"
+version = "2.21.0"
 description = "The official Python library for the together API"
 dynamic = ["readme"]
 license = "Apache-2.0"

src/together/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "together"
-__version__ = "2.20.0"  # x-release-please-version
+__version__ = "2.21.0"  # x-release-please-version

src/together/resources/beta/jig/queue.py

@@ -17,7 +17,14 @@
     async_to_streamed_response_wrapper,
 )
 from ...._base_client import make_request_options
-from ....types.beta.jig import queue_cancel_params, queue_submit_params, queue_metrics_params, queue_retrieve_params
+from ....types.beta.jig import (
+    queue_clear_params,
+    queue_cancel_params,
+    queue_submit_params,
+    queue_metrics_params,
+    queue_retrieve_params,
+)
+from ....types.beta.jig.queue_clear_response import QueueClearResponse
 from ....types.beta.jig.queue_cancel_response import QueueCancelResponse
 from ....types.beta.jig.queue_submit_response import QueueSubmitResponse
 from ....types.beta.jig.queue_metrics_response import QueueMetricsResponse
@@ -140,6 +147,42 @@ def cancel(
             cast_to=QueueCancelResponse,
         )
 
+    def clear(
+        self,
+        *,
+        model: str,
+        # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
+        # The extra values given here take precedence over values defined on the client or passed to this method.
+        extra_headers: Headers | None = None,
+        extra_query: Query | None = None,
+        extra_body: Body | None = None,
+        timeout: float | httpx.Timeout | None | NotGiven = not_given,
+    ) -> QueueClearResponse:
+        """Cancel all pending jobs for the given model.
+
+        Running jobs are left untouched.
+        Returns the number of jobs that were canceled.
+
+        Args:
+          model: Model identifier whose pending jobs should be canceled
+
+          extra_headers: Send extra headers
+
+          extra_query: Add additional query parameters to the request
+
+          extra_body: Add additional JSON properties to the request
+
+          timeout: Override the client-level default timeout for this request, in seconds
+        """
+        return self._post(
+            "/queue/clear",
+            body=maybe_transform({"model": model}, queue_clear_params.QueueClearParams),
+            options=make_request_options(
+                extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
+            ),
+            cast_to=QueueClearResponse,
+        )
+
     def metrics(
         self,
         *,
@@ -350,6 +393,42 @@ async def cancel(
             cast_to=QueueCancelResponse,
         )
 
+    async def clear(
+        self,
+        *,
+        model: str,
+        # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
+        # The extra values given here take precedence over values defined on the client or passed to this method.
+        extra_headers: Headers | None = None,
+        extra_query: Query | None = None,
+        extra_body: Body | None = None,
+        timeout: float | httpx.Timeout | None | NotGiven = not_given,
+    ) -> QueueClearResponse:
+        """Cancel all pending jobs for the given model.
+
+        Running jobs are left untouched.
+        Returns the number of jobs that were canceled.
+
+        Args:
+          model: Model identifier whose pending jobs should be canceled
+
+          extra_headers: Send extra headers
+
+          extra_query: Add additional query parameters to the request
+
+          extra_body: Add additional JSON properties to the request
+
+          timeout: Override the client-level default timeout for this request, in seconds
+        """
+        return await self._post(
+            "/queue/clear",
+            body=await async_maybe_transform({"model": model}, queue_clear_params.QueueClearParams),
+            options=make_request_options(
+                extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
+            ),
+            cast_to=QueueClearResponse,
+        )
+
     async def metrics(
         self,
         *,
@@ -456,6 +535,9 @@ def __init__(self, queue: QueueResource) -> None:
         self.cancel = to_raw_response_wrapper(
             queue.cancel,
         )
+        self.clear = to_raw_response_wrapper(
+            queue.clear,
+        )
         self.metrics = to_raw_response_wrapper(
             queue.metrics,
         )
@@ -474,6 +556,9 @@ def __init__(self, queue: AsyncQueueResource) -> None:
         self.cancel = async_to_raw_response_wrapper(
             queue.cancel,
         )
+        self.clear = async_to_raw_response_wrapper(
+            queue.clear,
+        )
         self.metrics = async_to_raw_response_wrapper(
             queue.metrics,
         )
@@ -492,6 +577,9 @@ def __init__(self, queue: QueueResource) -> None:
         self.cancel = to_streamed_response_wrapper(
             queue.cancel,
         )
+        self.clear = to_streamed_response_wrapper(
+            queue.clear,
+        )
         self.metrics = to_streamed_response_wrapper(
             queue.metrics,
         )
@@ -510,6 +598,9 @@ def __init__(self, queue: AsyncQueueResource) -> None:
         self.cancel = async_to_streamed_response_wrapper(
             queue.cancel,
         )
+        self.clear = async_to_streamed_response_wrapper(
+            queue.clear,
+        )
         self.metrics = async_to_streamed_response_wrapper(
             queue.metrics,
         )

src/together/resources/fine_tuning.py

@@ -115,6 +115,10 @@ def create(
         wandb_name: str | None = None,
         wandb_entity: str | None = None,
         random_seed: int | None = None,
+        early_stopping_enabled: bool = False,
+        early_stopping_patience: int | None = None,
+        early_stopping_min_delta: float | None = None,
+        early_stopping_warmup_evals: int | None = None,
         verbose: bool = False,
         model_limits: FinetuneTrainingLimits | None = None,
         train_on_inputs: bool | Literal["auto"] | None = None,
@@ -173,6 +177,15 @@ def create(
             random_seed (int, optional): Random seed for reproducible training (e.g. 42). When set, the same seed produces
                 the same run (e.g. data shuffle, init). If not provided (None), the server uses its default seed (42).
                 Defaults to None.
+            early_stopping_enabled (bool, optional): Stop training early when validation eval_loss stops improving.
+                Requires a validation_file, and n_evals must be at least
+                early_stopping_patience + early_stopping_warmup_evals + 1. Defaults to False.
+            early_stopping_patience (int, optional): Consecutive non-improving evals to tolerate before stopping.
+                Only applies when early_stopping_enabled is True. Defaults to None (server default 2).
+            early_stopping_min_delta (float, optional): Minimum eval_loss decrease to count as an improvement.
+                Only applies when early_stopping_enabled is True. Defaults to None (server default 0.0).
+            early_stopping_warmup_evals (int, optional): Initial evals to skip before counting patience.
+                Only applies when early_stopping_enabled is True. Defaults to None (server default 1).
             verbose (bool, optional): whether to print the job parameters before submitting a request.
                 Defaults to False.
             model_limits (FinetuneTrainingLimits, optional): Limits for the hyperparameters the model in Fine-tuning.
@@ -248,6 +261,10 @@ def create(
             wandb_name=wandb_name,
             wandb_entity=wandb_entity,
             random_seed=random_seed,
+            early_stopping_enabled=early_stopping_enabled,
+            early_stopping_patience=early_stopping_patience,
+            early_stopping_min_delta=early_stopping_min_delta,
+            early_stopping_warmup_evals=early_stopping_warmup_evals,
             train_on_inputs=train_on_inputs,
             training_method=training_method,
             dpo_beta=dpo_beta,
@@ -750,6 +767,10 @@ async def create(
         wandb_name: str | None = None,
         wandb_entity: str | None = None,
         random_seed: int | None = None,
+        early_stopping_enabled: bool = False,
+        early_stopping_patience: int | None = None,
+        early_stopping_min_delta: float | None = None,
+        early_stopping_warmup_evals: int | None = None,
         verbose: bool = False,
         model_limits: FinetuneTrainingLimits | None = None,
         train_on_inputs: bool | Literal["auto"] | None = None,
@@ -807,6 +828,15 @@ async def create(
             random_seed (int, optional): Random seed for reproducible training (e.g. 42). When set, the same seed produces
                 the same run (e.g. data shuffle, init). If not provided (None), the server uses its default seed (42).
                 Defaults to None.
+            early_stopping_enabled (bool, optional): Stop training early when validation eval_loss stops improving.
+                Requires a validation_file, and n_evals must be at least
+                early_stopping_patience + early_stopping_warmup_evals + 1. Defaults to False.
+            early_stopping_patience (int, optional): Consecutive non-improving evals to tolerate before stopping.
+                Only applies when early_stopping_enabled is True. Defaults to None (server default 2).
+            early_stopping_min_delta (float, optional): Minimum eval_loss decrease to count as an improvement.
+                Only applies when early_stopping_enabled is True. Defaults to None (server default 0.0).
+            early_stopping_warmup_evals (int, optional): Initial evals to skip before counting patience.
+                Only applies when early_stopping_enabled is True. Defaults to None (server default 1).
             verbose (bool, optional): whether to print the job parameters before submitting a request.
                 Defaults to False.
             model_limits (FinetuneTrainingLimits, optional): Limits for the hyperparameters the model in Fine-tuning.
@@ -882,6 +912,10 @@ async def create(
             wandb_name=wandb_name,
             wandb_entity=wandb_entity,
             random_seed=random_seed,
+            early_stopping_enabled=early_stopping_enabled,
+            early_stopping_patience=early_stopping_patience,
+            early_stopping_min_delta=early_stopping_min_delta,
+            early_stopping_warmup_evals=early_stopping_warmup_evals,
             train_on_inputs=train_on_inputs,
             training_method=training_method,
             dpo_beta=dpo_beta,

src/together/types/beta/cluster.py

@@ -13,9 +13,11 @@
     "AddOnConfig",
     "AddOnConfigDashboard",
     "AddOnConfigIngress",
+    "AddOnConfigTorchpass",
     "AddOnState",
     "AddOnStateDashboard",
     "AddOnStateIngress",
+    "AddOnStateTorchpass",
     "ControlPlaneNode",
     "ControlPlaneNodePhaseTransition",
     "GPUWorkerNode",
@@ -41,11 +43,23 @@ class AddOnConfigIngress(BaseModel):
     enabled: Optional[bool] = None
 
 
+class AddOnConfigTorchpass(BaseModel):
+    """Configuration for the Model Aware TorchPass add-on."""
+
+    enabled: Optional[bool] = None
+    """Whether to enable the Model Aware TorchPass add-on."""
+
+
 class AddOnConfig(BaseModel):
+    """Configuration for a cluster add-on."""
+
     dashboard: Optional[AddOnConfigDashboard] = None
 
     ingress: Optional[AddOnConfigIngress] = None
 
+    torchpass: Optional[AddOnConfigTorchpass] = None
+    """Configuration for the Model Aware TorchPass add-on."""
+
 
 class AddOnStateDashboard(BaseModel):
     pass
@@ -55,22 +69,35 @@ class AddOnStateIngress(BaseModel):
     pass
 
 
+class AddOnStateTorchpass(BaseModel):
+    """State for the Model Aware TorchPass add-on."""
+
+    pass
+
+
 class AddOnState(BaseModel):
+    """State for a cluster add-on."""
+
     dashboard: Optional[AddOnStateDashboard] = None
 
     ingress: Optional[AddOnStateIngress] = None
 
+    torchpass: Optional[AddOnStateTorchpass] = None
+    """State for the Model Aware TorchPass add-on."""
+
 
 class AddOn(BaseModel):
     """AddOnInfo is returned in cluster responses and add-on CRUD operations."""
 
     add_on_type: str
 
     config: AddOnConfig
+    """Configuration for a cluster add-on."""
 
     name: str
 
     state: AddOnState
+    """State for a cluster add-on."""
 
 
 class ControlPlaneNodePhaseTransition(BaseModel):
@@ -291,6 +318,12 @@ class ClusterConfig(BaseModel):
     init, extra conf).
     """
 
+    ssh_ca_enabled: Optional[bool] = None
+    """
+    Whether this cluster uses a per-cluster SSH certificate authority for
+    OIDC-signed SSH access.
+    """
+
 
 class DeletedGPUWorkerNodePhaseTransition(BaseModel):
     phase: Literal[