@meetkai/mka1 0.51.18 → 0.51.19

package/src/sdk/deployments.ts

@@ -0,0 +1,244 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { servingDeploymentsCreate } from "../funcs/servingDeploymentsCreate.js";
+import { servingDeploymentsDelete } from "../funcs/servingDeploymentsDelete.js";
+import { servingDeploymentsDeregister } from "../funcs/servingDeploymentsDeregister.js";
+import { servingDeploymentsGet } from "../funcs/servingDeploymentsGet.js";
+import { servingDeploymentsGetLogs } from "../funcs/servingDeploymentsGetLogs.js";
+import { servingDeploymentsGetMetrics } from "../funcs/servingDeploymentsGetMetrics.js";
+import { servingDeploymentsList } from "../funcs/servingDeploymentsList.js";
+import { servingDeploymentsListReplicas } from "../funcs/servingDeploymentsListReplicas.js";
+import { servingDeploymentsListRevisions } from "../funcs/servingDeploymentsListRevisions.js";
+import { servingDeploymentsRegister } from "../funcs/servingDeploymentsRegister.js";
+import { servingDeploymentsRollback } from "../funcs/servingDeploymentsRollback.js";
+import { servingDeploymentsUpdate } from "../funcs/servingDeploymentsUpdate.js";
+import { servingDeploymentsUpdateScaling } from "../funcs/servingDeploymentsUpdateScaling.js";
+import { ClientSDK, RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { unwrapAsync } from "../types/fp.js";
+
+export class Deployments extends ClientSDK {
+  /**
+   * List deployments
+   *
+   * @remarks
+   * List your inference deployments.
+   */
+  async list(
+    request: operations.ListDeploymentsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageDeployment> {
+    return unwrapAsync(servingDeploymentsList(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Create an inference deployment
+   *
+   * @remarks
+   * Create and start an inference deployment for a model, and return its endpoint.
+   */
+  async create(
+    request: operations.CreateDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<components.Deployment> {
+    return unwrapAsync(servingDeploymentsCreate(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Tear down a deployment
+   *
+   * @remarks
+   * Tear down a deployment and release its resources.
+   */
+  async delete(
+    request: operations.DeleteDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<void> {
+    return unwrapAsync(servingDeploymentsDelete(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Get a deployment
+   *
+   * @remarks
+   * Retrieve a single deployment, including its status and endpoint URL.
+   */
+  async get(
+    request: operations.GetDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<components.Deployment> {
+    return unwrapAsync(servingDeploymentsGet(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Update a deployment (new revision)
+   *
+   * @remarks
+   * Update a deployment's configuration, rolling out a new revision.
+   */
+  async update(
+    request: operations.UpdateDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<components.Deployment> {
+    return unwrapAsync(servingDeploymentsUpdate(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Fetch deployment logs
+   *
+   * @remarks
+   * Fetch recent log lines from a deployment's replicas.
+   */
+  async getLogs(
+    request: operations.GetDeploymentLogsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageLogLine> {
+    return unwrapAsync(servingDeploymentsGetLogs(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Live serving metrics
+   *
+   * @remarks
+   * Return live serving metrics (throughput, queue depth, accelerator usage).
+   */
+  async getMetrics(
+    request: operations.GetDeploymentMetricsRequest,
+    options?: RequestOptions,
+  ): Promise<components.Metrics> {
+    return unwrapAsync(servingDeploymentsGetMetrics(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Deregister the model from the LLM gateway
+   *
+   * @remarks
+   * Withdraw the deployment's model from the LLM gateway registry: inference stops resolving, compute keeps running (re-register any time).
+   */
+  async deregister(
+    request: operations.DeregisterDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<void> {
+    return unwrapAsync(servingDeploymentsDeregister(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Register the model with the LLM gateway
+   *
+   * @remarks
+   * Make the deployment's model available for inference: registers it in the LLM gateway's model registry under the deployment's name. Deployment must be ready. Registration is deliberately decoupled from deployment so pricing, budgets, and rate limits can be configured before the model becomes callable.
+   */
+  async register(
+    request: operations.RegisterDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<components.Deployment> {
+    return unwrapAsync(servingDeploymentsRegister(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * List running replicas
+   *
+   * @remarks
+   * List the replicas currently backing a deployment.
+   */
+  async listReplicas(
+    request: operations.ListReplicasRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageReplica> {
+    return unwrapAsync(servingDeploymentsListReplicas(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * List config revisions
+   *
+   * @remarks
+   * List a deployment's configuration revisions, most recent first.
+   */
+  async listRevisions(
+    request: operations.ListRevisionsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageRevision> {
+    return unwrapAsync(servingDeploymentsListRevisions(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Roll back to a previous revision
+   *
+   * @remarks
+   * Roll a deployment back to a previous configuration revision.
+   */
+  async rollback(
+    request: operations.RollbackDeploymentRequest,
+    options?: RequestOptions,
+  ): Promise<components.Deployment> {
+    return unwrapAsync(servingDeploymentsRollback(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Update autoscaling live (no redeploy)
+   *
+   * @remarks
+   * Change a deployment's autoscaling limits in place, without redeploying.
+   */
+  async updateScaling(
+    request: operations.UpdateScalingRequest,
+    options?: RequestOptions,
+  ): Promise<components.Deployment> {
+    return unwrapAsync(servingDeploymentsUpdateScaling(
+      this,
+      request,
+      options,
+    ));
+  }
+}

package/src/sdk/eligibility.ts

@@ -0,0 +1,28 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { servingEligibilityPreview } from "../funcs/servingEligibilityPreview.js";
+import { ClientSDK, RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { unwrapAsync } from "../types/fp.js";
+
+export class Eligibility extends ClientSDK {
+  /**
+   * Preview accelerator eligibility
+   *
+   * @remarks
+   * Preview, before submitting a job or deployment, how much matching capacity the cluster can currently acquire and how likely it is to start. The response is provider-anonymous: counts, price bands, and advice only. This is an advisory read; it reserves nothing.
+   */
+  async preview(
+    request: operations.PreviewEligibilityRequest,
+    options?: RequestOptions,
+  ): Promise<components.EligibilityPreview> {
+    return unwrapAsync(servingEligibilityPreview(
+      this,
+      request,
+      options,
+    ));
+  }
+}

package/src/sdk/finetunejobs.ts

@@ -0,0 +1,154 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { servingFinetuneJobsCancel } from "../funcs/servingFinetuneJobsCancel.js";
+import { servingFinetuneJobsCreate } from "../funcs/servingFinetuneJobsCreate.js";
+import { servingFinetuneJobsGet } from "../funcs/servingFinetuneJobsGet.js";
+import { servingFinetuneJobsGetEvents } from "../funcs/servingFinetuneJobsGetEvents.js";
+import { servingFinetuneJobsGetLogs } from "../funcs/servingFinetuneJobsGetLogs.js";
+import { servingFinetuneJobsGetMetrics } from "../funcs/servingFinetuneJobsGetMetrics.js";
+import { servingFinetuneJobsList } from "../funcs/servingFinetuneJobsList.js";
+import { servingFinetuneJobsListCheckpoints } from "../funcs/servingFinetuneJobsListCheckpoints.js";
+import { ClientSDK, RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { unwrapAsync } from "../types/fp.js";
+
+export class FinetuneJobs extends ClientSDK {
+  /**
+   * List fine-tune jobs
+   *
+   * @remarks
+   * List your fine-tune jobs.
+   */
+  async list(
+    request: operations.ListFinetuneJobsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageFinetuneJob> {
+    return unwrapAsync(servingFinetuneJobsList(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Submit a fine-tune job
+   *
+   * @remarks
+   * Submit a fine-tune job for a base model and dataset, and return its id.
+   */
+  async create(
+    request: operations.CreateFinetuneJobRequest,
+    options?: RequestOptions,
+  ): Promise<components.FinetuneJob> {
+    return unwrapAsync(servingFinetuneJobsCreate(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Get a fine-tune job
+   *
+   * @remarks
+   * Retrieve a fine-tune job, including its status and progress.
+   */
+  async get(
+    request: operations.GetFinetuneJobRequest,
+    options?: RequestOptions,
+  ): Promise<components.FinetuneJob> {
+    return unwrapAsync(servingFinetuneJobsGet(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Cancel a fine-tune job
+   *
+   * @remarks
+   * Cancel a running fine-tune job and return the updated job. Fine-tune records are never deleted: the cancelled job, its events, and its spend history stay readable.
+   */
+  async cancel(
+    request: operations.CancelFinetuneJobRequest,
+    options?: RequestOptions,
+  ): Promise<components.FinetuneJob> {
+    return unwrapAsync(servingFinetuneJobsCancel(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * List output checkpoints
+   *
+   * @remarks
+   * List the checkpoints a fine-tune job has produced.
+   */
+  async listCheckpoints(
+    request: operations.ListFinetuneCheckpointsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageCheckpoint> {
+    return unwrapAsync(servingFinetuneJobsListCheckpoints(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Status timeline
+   *
+   * @remarks
+   * Return the job's status timeline (queued, running, checkpoints, completion).
+   */
+  async getEvents(
+    request: operations.GetFinetuneEventsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageJobEvent> {
+    return unwrapAsync(servingFinetuneJobsGetEvents(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Fetch training logs
+   *
+   * @remarks
+   * Fetch recent training log lines for a job.
+   */
+  async getLogs(
+    request: operations.GetFinetuneLogsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageLogLine> {
+    return unwrapAsync(servingFinetuneJobsGetLogs(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Training metrics
+   *
+   * @remarks
+   * Return the training metric series (loss, learning rate, ...) pushed by the training run.
+   */
+  async getMetrics(
+    request: operations.GetFinetuneMetricsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageTrainingMetricPoint> {
+    return unwrapAsync(servingFinetuneJobsGetMetrics(
+      this,
+      request,
+      options,
+    ));
+  }
+}

package/src/sdk/recipes.ts

@@ -0,0 +1,100 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { servingRecipesDelete } from "../funcs/servingRecipesDelete.js";
+import { servingRecipesGet } from "../funcs/servingRecipesGet.js";
+import { servingRecipesList } from "../funcs/servingRecipesList.js";
+import { servingRecipesRegister } from "../funcs/servingRecipesRegister.js";
+import { servingRecipesSetEnabled } from "../funcs/servingRecipesSetEnabled.js";
+import { ClientSDK, RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { unwrapAsync } from "../types/fp.js";
+
+export class Recipes extends ClientSDK {
+  /**
+   * List serving recipes
+   *
+   * @remarks
+   * List the cluster-global serving recipe catalog. Tenants see only enabled (deployable) recipes; cluster administrators see every registered recipe including disabled ones.
+   */
+  async list(
+    request: operations.ListServingRecipesRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageServingRecipe> {
+    return unwrapAsync(servingRecipesList(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Delete a serving recipe
+   *
+   * @remarks
+   * Remove a recipe from the catalog. Rejected while any live deployment's active revision references the recipe. Cluster administrators only.
+   */
+  async delete(
+    request: operations.DeleteServingRecipeRequest,
+    options?: RequestOptions,
+  ): Promise<void> {
+    return unwrapAsync(servingRecipesDelete(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Get a serving recipe
+   *
+   * @remarks
+   * Get one recipe catalog record. Tenants can read enabled recipes; disabled recipes are visible to cluster administrators only.
+   */
+  async get(
+    request: operations.GetServingRecipeRequest,
+    options?: RequestOptions,
+  ): Promise<components.ServingRecipe> {
+    return unwrapAsync(servingRecipesGet(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Enable or disable a serving recipe
+   *
+   * @remarks
+   * Toggle cluster-wide deployability of a registered recipe. Disabling does not stop deployments already running the recipe. Cluster administrators only.
+   */
+  async setEnabled(
+    request: operations.SetServingRecipeEnabledRequest,
+    options?: RequestOptions,
+  ): Promise<components.ServingRecipe> {
+    return unwrapAsync(servingRecipesSetEnabled(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Register a serving recipe
+   *
+   * @remarks
+   * Register (or re-register) a digest-pinned recipe manifest in the catalog. A new recipe registers disabled; re-registration replaces the manifest and preserves enablement. Cluster administrators only.
+   */
+  async register(
+    request: operations.RegisterServingRecipeRequest,
+    options?: RequestOptions,
+  ): Promise<components.ServingRecipe> {
+    return unwrapAsync(servingRecipesRegister(
+      this,
+      request,
+      options,
+    ));
+  }
+}

package/src/sdk/sdk.ts

@@ -14,6 +14,7 @@ import { Guardrails } from "./guardrails.js";
 import { Llm } from "./llm.js";
 import { Permissions } from "./permissions.js";
 import { Sandbox } from "./sandbox.js";
+import { Serving } from "./serving.js";
 import { Usage } from "./usage.js";
 
 export class SDK extends ClientSDK {
@@ -76,4 +77,9 @@ export class SDK extends ClientSDK {
   get budgets(): Budgets {
     return (this._budgets ??= new Budgets(this._options));
   }
+
+  private _serving?: Serving;
+  get serving(): Serving {
+    return (this._serving ??= new Serving(this._options));
+  }
 }

package/src/sdk/sdkimages.ts

@@ -0,0 +1,82 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { servingImagesDelete } from "../funcs/servingImagesDelete.js";
+import { servingImagesGet } from "../funcs/servingImagesGet.js";
+import { servingImagesList } from "../funcs/servingImagesList.js";
+import { servingImagesRegister } from "../funcs/servingImagesRegister.js";
+import { ClientSDK, RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { unwrapAsync } from "../types/fp.js";
+
+export class SDKImages extends ClientSDK {
+  /**
+   * List images
+   *
+   * @remarks
+   * List your container images.
+   */
+  async list(
+    request: operations.ListImagesRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageImage> {
+    return unwrapAsync(servingImagesList(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Register a container image
+   *
+   * @remarks
+   * Register a container image by reference so it can back a BYO deployment or a recipe. (Data upload and build-with-weight-baking are future additions.)
+   */
+  async register(
+    request: operations.RegisterImageRequest,
+    options?: RequestOptions,
+  ): Promise<components.Image> {
+    return unwrapAsync(servingImagesRegister(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Delete an image
+   *
+   * @remarks
+   * Delete a container image.
+   */
+  async delete(
+    request: operations.DeleteImageRequest,
+    options?: RequestOptions,
+  ): Promise<void> {
+    return unwrapAsync(servingImagesDelete(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Get an image
+   *
+   * @remarks
+   * Retrieve an image, including its build status.
+   */
+  async get(
+    request: operations.GetImageRequest,
+    options?: RequestOptions,
+  ): Promise<components.Image> {
+    return unwrapAsync(servingImagesGet(
+      this,
+      request,
+      options,
+    ));
+  }
+}

package/src/sdk/sdkmodels.ts

@@ -0,0 +1,82 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { servingModelsDelete } from "../funcs/servingModelsDelete.js";
+import { servingModelsGet } from "../funcs/servingModelsGet.js";
+import { servingModelsList } from "../funcs/servingModelsList.js";
+import { servingModelsRegister } from "../funcs/servingModelsRegister.js";
+import { ClientSDK, RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { unwrapAsync } from "../types/fp.js";
+
+export class SDKModels extends ClientSDK {
+  /**
+   * List models
+   *
+   * @remarks
+   * List your registered models.
+   */
+  async list(
+    request: operations.ListServingModelsRequest,
+    options?: RequestOptions,
+  ): Promise<components.PageModel> {
+    return unwrapAsync(servingModelsList(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Register a model
+   *
+   * @remarks
+   * Register a model (from a source or volume) so it can be deployed or fine-tuned.
+   */
+  async register(
+    request: operations.RegisterServingModelRequest,
+    options?: RequestOptions,
+  ): Promise<components.Model> {
+    return unwrapAsync(servingModelsRegister(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Deregister a model
+   *
+   * @remarks
+   * Deregister a model from the catalog.
+   */
+  async delete(
+    request: operations.DeleteServingModelRequest,
+    options?: RequestOptions,
+  ): Promise<void> {
+    return unwrapAsync(servingModelsDelete(
+      this,
+      request,
+      options,
+    ));
+  }
+
+  /**
+   * Get a model
+   *
+   * @remarks
+   * Retrieve a registered model.
+   */
+  async get(
+    request: operations.GetServingModelRequest,
+    options?: RequestOptions,
+  ): Promise<components.Model> {
+    return unwrapAsync(servingModelsGet(
+      this,
+      request,
+      options,
+    ));
+  }
+}