runbios-sdk 0.5.0 → 0.6.0-rc.369

package/CHANGELOG.md

@@ -5,6 +5,29 @@ surface is added, renamed or removed, the patch for a fix inside an unchanged
 surface. The SDK, the Python SDK (`runbios-sdk` on PyPI) and the MCP server
 (`runbios-mcp`) are released together, at one version.
 
+## 0.6.0
+
+Added
+
+Managed training (the platform chooses the hardware; customers pay per trained token)
+
+- `managed_options()` / `get_managed_training_options`: the context lengths
+  (2K-128K), precisions (`bf16`, `fp8`) and optimizer states (16-bit, 8-bit)
+  each model and adapter is tested for.
+- Named inputs for those choices on create and preflight: `precision`,
+  `optim` and the validation row count (`precision`, `optim`, `validationRows`), mapped to
+  `config.precision`, `config.optim` and `config.split_dataset_samples`.
+- Under managed training, GPU type and count, GPU priorities, the queue
+  switch, the hourly price cap, per-device batch, accumulation and checkpoint
+  count are chosen by the platform and ignored if sent. A job with no free
+  machine waits in the queue automatically, unbilled, and can be cancelled.
+- Training logs and machine metrics (memory, MFU, throughput) are not
+  available to customers under managed training; training metrics are.
+
+Also
+
+- `training.liveDeployments` / `useLiveDeployment` for the Conscious Loop live-redeploy routes.
+
 ## 0.5.0
 
 Added

package/dist/client.js

@@ -431,7 +431,7 @@ export class HttpClient {
     constructor(config) {
         // Default host stays api.runbios.ai for now; cutover to api.runbios.ai is
         // planned once its DNS exists.
-        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api.runbios.ai').replace(/\/+$/, '');
+        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api-staging.runbios.ai').replace(/\/+$/, '');
         this.apiKey = config.apiKey ?? envApiKey();
         this.accessToken = config.accessToken;
         this.orgId = config.orgId;

package/dist/index.d.ts

@@ -36,7 +36,7 @@ import { Loop } from './resources/loop.js';
  * SDK version. Sent as part of the User-Agent header.
  * Must match package.json "version" -- enforced by a contract test.
  */
-export declare const VERSION = "0.5.0";
+export declare const VERSION = "0.6.0-rc.369";
 export declare class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     readonly models: Models;
@@ -79,7 +79,7 @@ export { Wallet } from './resources/wallet.js';
 export { GPU, type GPURecommendation } from './resources/gpu.js';
 export { LoopNoVersionError, type LoopChatParams, type LoopChatResult, } from './resources/loop-chat.js';
 export { Inference, validateChatRequest, parseSSE, buildInferenceRequest, contextSizingBasis, type ContextSizingBasis, CONTEXT_DEFAULT_CEILING, CONTEXT_EDITABLE_FLOOR, type ChatMessage, type FunctionTool, type ChatCompletionParams, type ChatCompletionResponse, type ChatCompletionChunk, type InferenceVerbParams, type CompletionParams, type EmbeddingParams, type RerankParams, type AnthropicMessageParams, type AnthropicMessageResponse, type LoopCaptureOption, type LoopSampleIdOnResult, LOOP_PIPELINE_HEADER, LOOP_SAMPLE_ID_HEADER, type ServerlessUsageWindow, type ServerlessTimeseriesMetric, type ServerlessWorkspaceLimits, type ServerlessUsageEnvelope, type ServerlessUsageOverviewResponse, type ServerlessUsageRowsResponse, type ServerlessUsageTimeseriesResponse, type ServerlessUsageDailyResponse, type ServerlessSavingsTotals, type ServerlessSavingsPeriod, type ServerlessUsageSavingsResponse, } from './resources/inference.js';
-export type { BiOSConfig, PaginatedResponse, ApiErrorBody, AvailableGpuAlternative, CapacityMinimumRequirement, InferenceBookingAccepted, Model, ModelSearchParams, ModelSearchResponse, ModelDetailResponse, InferenceModel, InferenceModelListResponse, ModelConfig, Dataset, DatasetListParams, DatasetListResponse, DatasetUploadParams, DatasetPreview, DatasetPreviewParams, DatasetImportHFParams, DatasetRegisterHFParams, DatasetHubSearchParams, DatasetHubPreviewParams, DatasetValidation, DatasetFormatVariant, DatasetFormatSpec, DatasetFormatSpecs, DatasetStorageUsage, TrainingMethod, RLHFAlgorithm, AdapterType, TrainingJobStatus, TrainingStatusFilter, TrainingCreateParams, TrainingStartFromParams, TrainingStartFrom, TrainingExportStep, TrainingListParams, TrainingListResponse, TrainingJob, TrainingMetrics, TrainingResourceMetrics, TrainingDeviceMetrics, MetricPoint, MetricGraphConfig, TrainingCheckpoint, TrainingLogs, TrainingLogEntry, TrainingStopResponse, TrainingResumeResponse, CanonicalTrainingRequest, TrainingPreflightDataset, TrainingPreflightWarning, TrainingPreflightResponse, TrainingCapabilityChoice, TrainingConfigFieldCapability, TrainingCapabilities, TrainingFeature, GPUChoice, WalletBalance, Transaction, TransactionListResponse, TransactionListParams, GPUInfo, GPUPricingResponse, GPUOptionsParams, TrainingRecommendParams, TrainingAdvisorVerdict, TrainingAdvisorRecommendation, TrainingAdvisorUserShape, TrainingAdvisorJustification, TrainingAdvisorBasis, GPUOption, GPUOptionSuggestion, GPUOptionsResponse, InferenceStatus, InferenceCreateParams, InferenceUpdateParams, InferenceDeployment, InferenceDeploymentSummary, InferenceCreateResponse, InferenceListResponse, InferenceUpdateResponse, InferenceLifecycleResponse, InferenceDeleteResponse, InferenceGPUOptionMarket, InferenceGPUOption, InferenceGPUAlternative, InferenceGPUOptionsParams, InferenceGPUOptionsResponse, InferenceCanonicalRequest, InferencePreflightWarning, InferencePreflightResponse, AdapterCompatibility, AdapterCompatibilityResponse, AdapterCompatibilityParams, SupportedArchitecture, ArchitectureScope, SupportedArchitecturesResponse, ApiKey, ApiKeyScope, ApiKeyIntrospection, Organization, OrgMember, OrgInvite, Workspace, Integration, IntegrationCreateParams, StorageUsage, StorageObject, PromptTokensDetails, ChatCompletionUsage, ServingKind, TrainingRulePausedReason, TrainingTrigger, TrainingRecipe, TrainingRunState, TrainingVerdict, TrainingDecision, EvaluationStatus, EvaluationItemStatus, ItemErrorCode, EvaluationWinner, EvaluationItemWinnerFilter, ReportStatusFilter, ReportSort, EvaluationWarningCode, TrainingToolCall, TrainingMessage, TrainingRuleServing, TrainingRunPromoteRequest, TrainingRunRejectRequest, TrainingRunRollbackRequest, TrainingRunCancelRequest, TrainingRunListParams, TrainingRun, TrainingRunSummary, TrainingRunEvent, TrainingRunLinks, TrainingRunActions, EvaluationRef, EvaluationDecoding, EvaluationJudgeDimension, EvaluationDimensionScore, EvaluationWarning, EvaluationMargin, Evaluation, EvaluationItemJudge, EvaluationReferenceKind, EvaluationReference, EvaluationPairwiseOrder, EvaluationItemPairwise, EvaluationItemSide, EvaluationItem, EvaluationItemListParams, BenchmarkStatus, BenchmarkSourceKind, BenchmarkRunStatus, BenchmarkScoring, BenchmarkCheck, BenchmarkDimensionScore, Benchmark, BenchmarkItem, BenchmarkRun, BenchmarkRunItemSide, BenchmarkRunItem, BenchmarkRunItemListParams, BenchmarkHistoryPoint, BenchmarkSource, BenchmarkCreateParams, BenchmarkListParams, BenchmarkItemListParams, BenchmarkHistoryParams, BenchmarkRetireParams, InferenceAlias, InferenceAliasRequest, TrainingRuleDeleteResponse, TrainingRunListResponse, TrainingRunBenchmarkRun, TrainingRunResponse, TrainingRunActionResponse, EvaluationItemsResponse, BenchmarkRunItemsResponse, InferenceAliasListResponse, InferenceAliasResponse, InferenceAliasDeleteResponse, BenchmarkListResponse, BenchmarkItemsResponse, BenchmarkHistoryResponse, BenchmarkQuestionsParams, BenchmarkQuestionsTraining, BenchmarkQuestionResult, BenchmarkQuestion, BenchmarkQuestionsResponse, TrainingCurvePoint, TrainingCurve, TrainingRuleRunRefusalCode, TrainingRuleSaveRefusalCode, TrainingRuleMonthlyLimitRefusal, PipelineStatus, PipelineActiveRun, PipelineVersion, PipelineAttempt, AttemptExplanation, AttemptMissedBar, PipelineModelRole, PipelineComparedWith, PipelineTotals, PipelineTrainingsByResult, PipelineTrainingsByModel, PipelineSpentParts, PipelineAttemptListParams, PipelineAttemptListResponse, PipelineAttemptModelFilter, PipelineAttemptResultFilter, PipelineAttemptTrigger, PipelineSecondModel, PipelineSecondModelState, Lineage, PipelineChain, PipelineChainRebuild, TrainingRunRollbackRefusalCode, TrainingRunRetestRefusalCode, LoopTraceRebuild, LoopTraceDeleteResponse, PipelineAttemptOutcome, PipelineLiveDeploymentStatus, PipelineLiveDeployment, LoopStorageWhere, LoopStorage, Pipeline, PipelineListParams, PipelineListStatus, PipelineListResponse, PipelineTrainingSettings, PipelineTrainingDefaults, PipelineTrainingRequest, PipelineResponse, PipelineTrainWhen, PipelineSchedule, PipelineWriteRequest, PipelineMutationResponse, PipelineImportParams, PipelineFloors, HeldBackFloors, LoopModel, LoopModelsResponse, LoopUnavailableModel, LoopUnavailableReasonCode, LoopTracePipeline, LoopTracePipelinesResponse, LoopMetricsHistoryParams, LoopMetricPoint, LoopSignalDay, LoopMetricsHistory, LoopCaptureFeedback, LoopFeedbackWord, LoopCaptureHealth, LoopCaptureMiss, LoopSignalManyItem, LoopSignalManyResult, PromotionPolicyState, RuleBenchmark, PipelineBenchmarksResponse, PipelineBenchmarkAttachParams, PipelineBenchmarkMoveParams, CompareBenchmark, VersionBenchmarkScore, PromotionMeasurement, PromotionOutcome, VersionHoldout, VersionCommonSet, VersionScores, VersionScoreDifference, VersionCompareResponse, VersionCompareParams, TrainingRunPromotion, VerdictCheck, VerdictCheckKey, VerdictChecksWithheld, TrainingRunCost, AttemptDirection, AttemptHoldout, AttemptCompareSide, AttemptSharedSample, AttemptSharedFigures, AttemptBenchmarkSide, AttemptCompareBenchmark, AttemptCompareItemSide, AttemptCompareItem, AttemptCompareResponse, AttemptCompareParams, LoopTraceJudge, LoopJudgeFilter, LoopSampleState, LoopSampleStateName, LoopSampleDay, LoopImportRefusal, PipelineAutoScoring, PipelineAutoScoringStatus, LoopDatasetFeedback, LoopSampleOrigin, LoopFeedbackCounts, LoopPipelineCounts, LoopDatasetPipelineRef, LoopDataset, LoopDatasetWrite, LoopDatasetCreate, LoopDatasetPreviewParams, LoopDatasetPreviewModel, LoopDatasetPreview, LoopDatasetListParams, LoopDatasetListResponse, LoopDatasetCreator, LoopDatasetLearnedBy, LoopDatasetExportParams, LoopDatasetExportLine, LoopDatasetExport, LoopDatasetDeleteResponse, PipelineDatasetRef, PipelineConditions, PipelineConditionsWrite, TrainedOnDataset, TrainedOn, } from './types.js';
+export type { BiOSConfig, PaginatedResponse, ApiErrorBody, AvailableGpuAlternative, CapacityMinimumRequirement, InferenceBookingAccepted, Model, ModelSearchParams, ModelSearchResponse, ModelDetailResponse, InferenceModel, InferenceModelListResponse, ModelConfig, Dataset, DatasetListParams, DatasetListResponse, DatasetUploadParams, DatasetPreview, DatasetPreviewParams, DatasetImportHFParams, DatasetRegisterHFParams, DatasetHubSearchParams, DatasetHubPreviewParams, DatasetValidation, DatasetFormatVariant, DatasetFormatSpec, DatasetFormatSpecs, DatasetStorageUsage, TrainingMethod, RLHFAlgorithm, AdapterType, TrainingJobStatus, TrainingStatusFilter, TrainingCreateParams, TrainingStartFromParams, TrainingStartFrom, TrainingExportStep, TrainingListParams, TrainingListResponse, TrainingJob, TrainingMetrics, TrainingResourceMetrics, TrainingDeviceMetrics, MetricPoint, MetricGraphConfig, TrainingCheckpoint, TrainingLogs, TrainingLogEntry, TrainingStopResponse, TrainingResumeResponse, CanonicalTrainingRequest, TrainingPreflightDataset, TrainingPreflightWarning, TrainingPreflightResponse, TrainingCostEstimate, ManagedTrainingOption, ManagedTrainingOptions, ManagedTrainingOptionsParams, TrainingCapabilityChoice, TrainingConfigFieldCapability, TrainingCapabilities, TrainingFeature, GPUChoice, WalletBalance, Transaction, TransactionListResponse, TransactionListParams, GPUInfo, GPUPricingResponse, GPUOptionsParams, TrainingRecommendParams, TrainingAdvisorVerdict, TrainingAdvisorRecommendation, TrainingAdvisorUserShape, TrainingAdvisorJustification, TrainingAdvisorBasis, GPUOption, GPUOptionSuggestion, GPUOptionsResponse, InferenceStatus, InferenceCreateParams, InferenceUpdateParams, InferenceDeployment, InferenceDeploymentSummary, InferenceCreateResponse, InferenceListResponse, InferenceUpdateResponse, InferenceLifecycleResponse, InferenceDeleteResponse, InferenceGPUOptionMarket, InferenceGPUOption, InferenceGPUAlternative, InferenceGPUOptionsParams, InferenceGPUOptionsResponse, InferenceCanonicalRequest, InferencePreflightWarning, InferencePreflightResponse, AdapterCompatibility, AdapterCompatibilityResponse, AdapterCompatibilityParams, SupportedArchitecture, ArchitectureScope, SupportedArchitecturesResponse, ApiKey, ApiKeyScope, ApiKeyIntrospection, Organization, OrgMember, OrgInvite, Workspace, Integration, IntegrationCreateParams, StorageUsage, StorageObject, PromptTokensDetails, ChatCompletionUsage, ServingKind, TrainingRulePausedReason, TrainingTrigger, TrainingRecipe, TrainingRunState, TrainingVerdict, TrainingDecision, EvaluationStatus, EvaluationItemStatus, ItemErrorCode, EvaluationWinner, EvaluationItemWinnerFilter, ReportStatusFilter, ReportSort, EvaluationWarningCode, TrainingToolCall, TrainingMessage, TrainingRuleServing, TrainingRunPromoteRequest, TrainingRunRejectRequest, TrainingRunRollbackRequest, TrainingRunCancelRequest, TrainingRunListParams, TrainingRun, TrainingRunSummary, TrainingRunEvent, TrainingRunLinks, TrainingRunActions, EvaluationRef, EvaluationDecoding, EvaluationJudgeDimension, EvaluationDimensionScore, EvaluationWarning, EvaluationMargin, Evaluation, EvaluationItemJudge, EvaluationReferenceKind, EvaluationReference, EvaluationPairwiseOrder, EvaluationItemPairwise, EvaluationItemSide, EvaluationItem, EvaluationItemListParams, BenchmarkStatus, BenchmarkSourceKind, BenchmarkRunStatus, BenchmarkScoring, BenchmarkCheck, BenchmarkDimensionScore, Benchmark, BenchmarkItem, BenchmarkRun, BenchmarkRunItemSide, BenchmarkRunItem, BenchmarkRunItemListParams, BenchmarkHistoryPoint, BenchmarkSource, BenchmarkCreateParams, BenchmarkListParams, BenchmarkItemListParams, BenchmarkHistoryParams, BenchmarkRetireParams, InferenceAlias, InferenceAliasRequest, TrainingRuleDeleteResponse, TrainingRunListResponse, TrainingRunBenchmarkRun, TrainingRunResponse, TrainingRunActionResponse, EvaluationItemsResponse, BenchmarkRunItemsResponse, InferenceAliasListResponse, InferenceAliasResponse, InferenceAliasDeleteResponse, BenchmarkListResponse, BenchmarkItemsResponse, BenchmarkHistoryResponse, BenchmarkQuestionsParams, BenchmarkQuestionsTraining, BenchmarkQuestionResult, BenchmarkQuestion, BenchmarkQuestionsResponse, TrainingCurvePoint, TrainingCurve, TrainingRuleRunRefusalCode, TrainingRuleSaveRefusalCode, TrainingRuleMonthlyLimitRefusal, PipelineStatus, PipelineActiveRun, PipelineVersion, PipelineAttempt, AttemptExplanation, AttemptMissedBar, PipelineModelRole, PipelineComparedWith, PipelineTotals, PipelineTrainingsByResult, PipelineTrainingsByModel, PipelineSpentParts, PipelineAttemptListParams, PipelineAttemptListResponse, PipelineAttemptModelFilter, PipelineAttemptResultFilter, PipelineAttemptTrigger, PipelineSecondModel, PipelineSecondModelState, Lineage, PipelineChain, PipelineChainRebuild, TrainingRunRollbackRefusalCode, TrainingRunRetestRefusalCode, LoopTraceRebuild, LoopTraceDeleteResponse, PipelineAttemptOutcome, PipelineLiveDeploymentStatus, PipelineLiveDeployment, LoopStorageWhere, LoopStorage, Pipeline, PipelineListParams, PipelineListStatus, PipelineListResponse, PipelineTrainingSettings, PipelineTrainingDefaults, PipelineTrainingRequest, PipelineResponse, PipelineTrainWhen, PipelineSchedule, PipelineWriteRequest, PipelineMutationResponse, PipelineImportParams, PipelineFloors, HeldBackFloors, LoopModel, LoopModelsResponse, LoopUnavailableModel, LoopUnavailableReasonCode, LoopTracePipeline, LoopTracePipelinesResponse, LoopMetricsHistoryParams, LoopMetricPoint, LoopSignalDay, LoopMetricsHistory, LoopCaptureFeedback, LoopFeedbackWord, LoopCaptureHealth, LoopCaptureMiss, LoopSignalManyItem, LoopSignalManyResult, PromotionPolicyState, RuleBenchmark, PipelineBenchmarksResponse, PipelineBenchmarkAttachParams, PipelineBenchmarkMoveParams, CompareBenchmark, VersionBenchmarkScore, PromotionMeasurement, PromotionOutcome, VersionHoldout, VersionCommonSet, VersionScores, VersionScoreDifference, VersionCompareResponse, VersionCompareParams, TrainingRunPromotion, VerdictCheck, VerdictCheckKey, VerdictChecksWithheld, TrainingRunCost, AttemptDirection, AttemptHoldout, AttemptCompareSide, AttemptSharedSample, AttemptSharedFigures, AttemptBenchmarkSide, AttemptCompareBenchmark, AttemptCompareItemSide, AttemptCompareItem, AttemptCompareResponse, AttemptCompareParams, LoopTraceJudge, LoopJudgeFilter, LoopSampleState, LoopSampleStateName, LoopSampleDay, LoopImportRefusal, PipelineAutoScoring, PipelineAutoScoringStatus, LoopDatasetFeedback, LoopSampleOrigin, LoopFeedbackCounts, LoopPipelineCounts, LoopDatasetPipelineRef, LoopDataset, LoopDatasetWrite, LoopDatasetCreate, LoopDatasetPreviewParams, LoopDatasetPreviewModel, LoopDatasetPreview, LoopDatasetListParams, LoopDatasetListResponse, LoopDatasetCreator, LoopDatasetLearnedBy, LoopDatasetExportParams, LoopDatasetExportLine, LoopDatasetExport, LoopDatasetDeleteResponse, PipelineDatasetRef, PipelineConditions, PipelineConditionsWrite, TrainedOnDataset, TrainedOn, } from './types.js';
 /** The closed set of run states a training run never leaves. */
 export { TERMINAL_RUN_STATES } from './types.js';
 /** The most versions a pipeline may be set to make. */

package/dist/index.js

@@ -36,7 +36,7 @@ import { Loop } from './resources/loop.js';
  * SDK version. Sent as part of the User-Agent header.
  * Must match package.json "version" -- enforced by a contract test.
  */
-export const VERSION = '0.5.0';
+export const VERSION = '0.6.0-rc.369';
 export class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     models;

package/dist/resources/inference.js

@@ -373,7 +373,7 @@ export class Inference {
         this.key = config.inferenceKey || envInferenceKey() || envApiKey();
         // Same default host as HttpClient (api.runbios.ai); api.runbios.ai cutover
         // is planned once its DNS exists — update both call sites together.
-        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api.runbios.ai').replace(/\/+$/, '');
+        this.baseUrl = (config.baseUrl || envBaseUrl() || 'https://api-staging.runbios.ai').replace(/\/+$/, '');
         this.timeout = config.timeout ?? 900_000;
         this._http = http;
     }

package/dist/resources/loop.d.ts

@@ -598,6 +598,22 @@ export declare class Loop {
      * the base on every remaining sample, with no wait for new ones.
      */
     runPipeline(id: string): Promise<PipelineMutationResponse>;
+    /**
+     * The workspace's deployments serving exactly the live version's weights.
+     * For a live version whose deployment was deleted: deploy its weights again
+     * (`live.training_job_id`, `live.checkpoint_id`), then pick the deployment
+     * with {@link useLiveDeployment}. 409 `NO_LIVE_VERSION` when it has none.
+     */
+    liveDeployments(id: string): Promise<LiveDeploymentsResponse>;
+    /**
+     * Answer the pipeline's model name from this deployment of the live
+     * version's weights. Nothing is trained again; the deployment bills you by
+     * the hour, like any deployment. Refused while the pipeline's own
+     * deployment still exists (409 `LIVE_DEPLOYMENT_PRESENT`: start it again
+     * instead), for other weights (422 `DEPLOYMENT_WEIGHTS_DIFFER`) and for a
+     * deployment that is not running (409 `DEPLOYMENT_NOT_RUNNING`).
+     */
+    useLiveDeployment(id: string, deploymentId: string): Promise<UseLiveDeploymentResponse>;
     /**
      * {@link importRows}, with every row given the pipeline's own tag. `source`
      * defaults to that tag; the rows, the answer and the retry rules are the
@@ -851,3 +867,30 @@ export declare class Loop {
      */
     listBenchmarkRunItems(id: string, params?: BenchmarkRunItemListParams): Promise<BenchmarkRunItemsResponse>;
 }
+/** What {@link Loop.liveDeployments} answers. */
+export interface LiveDeploymentsResponse {
+    live: {
+        run_id: string;
+        version: number;
+        checkpoint_id: string;
+        training_job_id: string | null;
+    };
+    current_deployment_id: string | null;
+    deployments: Array<{
+        id: string;
+        name: string;
+        status: string;
+    }>;
+}
+/** What {@link Loop.useLiveDeployment} answers. */
+export interface UseLiveDeploymentResponse {
+    live: {
+        run_id: string;
+        version: number;
+        checkpoint_id: string;
+        training_job_id: string | null;
+    };
+    deployment_id: string;
+    status: string;
+    serving_name: string;
+}

package/dist/resources/loop.js

@@ -954,6 +954,26 @@ export class Loop {
     async runPipeline(id) {
         return this._http.fetchPost(`/api/loop/pipelines/${encodeURIComponent(id)}/run`, {});
     }
+    /**
+     * The workspace's deployments serving exactly the live version's weights.
+     * For a live version whose deployment was deleted: deploy its weights again
+     * (`live.training_job_id`, `live.checkpoint_id`), then pick the deployment
+     * with {@link useLiveDeployment}. 409 `NO_LIVE_VERSION` when it has none.
+     */
+    async liveDeployments(id) {
+        return this._http.fetchGet(`/api/loop/pipelines/${encodeURIComponent(id)}/live/deployments`);
+    }
+    /**
+     * Answer the pipeline's model name from this deployment of the live
+     * version's weights. Nothing is trained again; the deployment bills you by
+     * the hour, like any deployment. Refused while the pipeline's own
+     * deployment still exists (409 `LIVE_DEPLOYMENT_PRESENT`: start it again
+     * instead), for other weights (422 `DEPLOYMENT_WEIGHTS_DIFFER`) and for a
+     * deployment that is not running (409 `DEPLOYMENT_NOT_RUNNING`).
+     */
+    async useLiveDeployment(id, deploymentId) {
+        return this._http.fetchPost(`/api/loop/pipelines/${encodeURIComponent(id)}/live/deployment`, { deployment_id: deploymentId });
+    }
     /**
      * {@link importRows}, with every row given the pipeline's own tag. `source`
      * defaults to that tag; the rows, the answer and the retry rules are the

package/dist/resources/training.d.ts

@@ -1,5 +1,5 @@
 import type { HttpClient } from '../client.js';
-import type { TrainingCreateParams, TrainingListParams, TrainingJob, TrainingListResponse, TrainingMetrics, TrainingMetricsOptions, TrainingCheckpoint, TrainingLogs, TrainingStopResponse, TrainingResumeResponse, TrainingPreflightResponse, TrainingRecommendParams, TrainingAdvisorVerdict, TrainingCapabilities } from '../types.js';
+import type { TrainingCreateParams, TrainingListParams, TrainingJob, TrainingListResponse, TrainingMetrics, TrainingMetricsOptions, TrainingCheckpoint, TrainingLogs, TrainingStopResponse, TrainingResumeResponse, TrainingPreflightResponse, TrainingRecommendParams, TrainingAdvisorVerdict, TrainingCapabilities, ManagedTrainingOptions, ManagedTrainingOptionsParams } from '../types.js';
 /**
  * Create, monitor, and manage fine-tuning training jobs.
  */
@@ -35,6 +35,21 @@ export declare class Training {
      * when the alternatives list is empty; waiting is unbilled. The SDK never
      * auto-substitutes a GPU; a transient 503 is a retry, never a capacity verdict.
      *
+     * Managed training: when `capabilities().features` lists `managed_training`
+     * enabled, the platform chooses the hardware. Choose the context length,
+     * precision and optimizer state from `managedOptions()` (context 2K-128K,
+     * `bf16`/`fp8`, `adamw_torch_fused` 16-bit or `adamw_8bit` 8-bit state).
+     * `gpuType`, `gpuCount`, `gpuPriorities`, `queueIfUnavailable`,
+     * `maxPriceHourCents`, batch size, gradient accumulation and checkpoint
+     * count are then ignored (preflight warns `FIELD_IGNORED`). The schedule,
+     * validation rows (`split_dataset_samples`) and early stopping stay yours.
+     * Validation rows must be between 100 (or 5% of the rows when that is under
+     * 100) and 40% of the training rows; unset, 5% capped at 1,000 rows is held
+     * out unless the job has evaluation datasets, and preflight returns the
+     * exact rows and range in `validation`. The schedule defaults to cosine. It waits unbilled when no machine is free, keeps its best
+     * and latest checkpoints, and with `token_billing` is charged per trained
+     * token (`preflight` returns `estimate`).
+     *
      * @example
      * ```ts
      * const job = await client.training.create({
@@ -55,6 +70,14 @@ export declare class Training {
     create(params: TrainingCreateParams): Promise<TrainingJob>;
     /** Validate and canonicalize a training request without creating or billing a job. */
     preflight(params: TrainingCreateParams): Promise<TrainingPreflightResponse>;
+    /**
+     * Managed training: what this model, method and adapter can be trained with
+     * -- the context lengths, precisions (`bf16`, `fp8`) and optimizer states
+     * (`adamw_torch_fused` = 16-bit, `adamw_8bit` = 8-bit) its certified setups
+     * cover. The platform chooses the hardware; every option listed is accepted
+     * by `preflight` and `create`. Empty `options`: not trainable yet.
+     */
+    managedOptions(params: ManagedTrainingOptionsParams): Promise<ManagedTrainingOptions>;
     /**
      * Ask the trainer's own sizing model what to run: per-device batch,
      * accumulation, learning rate, warmup, predicted peak memory, minimum GPU
@@ -121,6 +144,11 @@ export declare class Training {
     /**
      * Get training logs for a job.
      *
+     * Not available when managed training is enabled (the server answers 404
+     * `LOGS_NOT_AVAILABLE`): progress is on {@link get} and training metrics on
+     * `getMetrics`, which then carry training metrics only (no memory, MFU or
+     * throughput).
+     *
      * @param id - Training job ID.
      * @param tail - Maximum newest entries to return (server clamps to 1-1000).
      *

package/dist/resources/training.js

@@ -160,6 +160,13 @@ function buildTrainingRequest(params) {
         config.early_stopping_patience = params.earlyStoppingPatience;
     if (params.overlongPolicy !== undefined)
         config.overlong_policy = params.overlongPolicy;
+    // Managed training choices (see managedOptions()).
+    if (params.precision !== undefined)
+        config.precision = params.precision;
+    if (params.optim !== undefined)
+        config.optim = params.optim;
+    if (params.validationRows !== undefined)
+        config.split_dataset_samples = params.validationRows;
     if (params.extraConfig)
         Object.assign(config, params.extraConfig);
     if (Object.keys(config).length > 0)
@@ -241,6 +248,21 @@ export class Training {
      * when the alternatives list is empty; waiting is unbilled. The SDK never
      * auto-substitutes a GPU; a transient 503 is a retry, never a capacity verdict.
      *
+     * Managed training: when `capabilities().features` lists `managed_training`
+     * enabled, the platform chooses the hardware. Choose the context length,
+     * precision and optimizer state from `managedOptions()` (context 2K-128K,
+     * `bf16`/`fp8`, `adamw_torch_fused` 16-bit or `adamw_8bit` 8-bit state).
+     * `gpuType`, `gpuCount`, `gpuPriorities`, `queueIfUnavailable`,
+     * `maxPriceHourCents`, batch size, gradient accumulation and checkpoint
+     * count are then ignored (preflight warns `FIELD_IGNORED`). The schedule,
+     * validation rows (`split_dataset_samples`) and early stopping stay yours.
+     * Validation rows must be between 100 (or 5% of the rows when that is under
+     * 100) and 40% of the training rows; unset, 5% capped at 1,000 rows is held
+     * out unless the job has evaluation datasets, and preflight returns the
+     * exact rows and range in `validation`. The schedule defaults to cosine. It waits unbilled when no machine is free, keeps its best
+     * and latest checkpoints, and with `token_billing` is charged per trained
+     * token (`preflight` returns `estimate`).
+     *
      * @example
      * ```ts
      * const job = await client.training.create({
@@ -280,6 +302,21 @@ export class Training {
         const { body } = buildTrainingRequest(params);
         return this._http.fetchPost('/api/training/preflight', body);
     }
+    /**
+     * Managed training: what this model, method and adapter can be trained with
+     * -- the context lengths, precisions (`bf16`, `fp8`) and optimizer states
+     * (`adamw_torch_fused` = 16-bit, `adamw_8bit` = 8-bit) its certified setups
+     * cover. The platform chooses the hardware; every option listed is accepted
+     * by `preflight` and `create`. Empty `options`: not trainable yet.
+     */
+    async managedOptions(params) {
+        const q = new URLSearchParams({
+            model_id: params.model,
+            training_method: params.method === 'cpt' ? 'pt' : params.method || 'sft',
+            train_type: params.adapter || 'lora',
+        });
+        return this._http.fetchGet(`/api/training/managed-options?${q}`);
+    }
     /**
      * Ask the trainer's own sizing model what to run: per-device batch,
      * accumulation, learning rate, warmup, predicted peak memory, minimum GPU
@@ -424,6 +461,11 @@ export class Training {
     /**
      * Get training logs for a job.
      *
+     * Not available when managed training is enabled (the server answers 404
+     * `LOGS_NOT_AVAILABLE`): progress is on {@link get} and training metrics on
+     * `getMetrics`, which then carry training metrics only (no memory, MFU or
+     * throughput).
+     *
      * @param id - Training job ID.
      * @param tail - Maximum newest entries to return (server clamps to 1-1000).
      *

package/dist/types.d.ts

@@ -8,13 +8,13 @@ export interface BiOSConfig {
     orgId?: string;
     /** Workspace ID. Optional when using API keys (resolved from the key). Can override for multi-workspace keys. */
     workspaceId?: string;
-    /** Base URL for the API. Falls back to the RUNBIOS_BASE_URL environment variable (legacy: BIOS_BASE_URL), then the canonical https://api.runbios.ai hostname. */
+    /** Base URL for the API. Falls back to the RUNBIOS_BASE_URL environment variable (legacy: BIOS_BASE_URL), then the canonical https://api-staging.runbios.ai hostname. */
     baseUrl?: string;
     /** Request timeout in milliseconds. Defaults to 30000. */
     timeout?: number;
     /** Default per-deployment inference key. Can be overridden per inference call. */
     inferenceKey?: string;
-    /** Inference base URL. Defaults to baseUrl, then https://api.runbios.ai. */
+    /** Inference base URL. Defaults to baseUrl, then https://api-staging.runbios.ai. */
     inferenceBaseUrl?: string;
     /** End-to-end inference timeout in milliseconds. Defaults to 15 minutes. */
     inferenceTimeout?: number;
@@ -730,6 +730,12 @@ export interface TrainingCreateParams {
     earlyStoppingPatience?: number;
     /** Samples longer than maxSeqLength: 'truncate' keeps the last tokens (default), 'drop' skips them. */
     overlongPolicy?: 'truncate' | 'drop';
+    /** Managed training: precision recipe from managedOptions() (bf16 or fp8). */
+    precision?: 'bf16' | 'fp8';
+    /** Managed training: optimizer state, adamw_torch_fused (16-bit) or adamw_8bit (8-bit). */
+    optim?: 'adamw_torch_fused' | 'adamw_8bit';
+    /** Managed training: validation rows held out and never trained on (0 turns validation off). */
+    validationRows?: number;
     /** Additional configuration passed directly to the training backend. */
     extraConfig?: Record<string, unknown>;
 }
@@ -1163,6 +1169,55 @@ export interface TrainingPreflightResponse {
      * you asked for is predicted to fit. Absent when no advisor is deployed.
      */
     advisor?: TrainingAdvisorVerdict;
+    /**
+     * Token and price estimate of a managed, token-billed run (present when the
+     * capabilities list `token_billing` enabled). You are charged the exact
+     * tokens trained; this is the estimate before launch.
+     */
+    estimate?: TrainingCostEstimate;
+}
+/** Estimated tokens and price of a managed training run. */
+/** One precision + optimizer-state combination a managed job may choose for a
+ * model, method and adapter, with the context lengths a certified setup covers
+ * for it. Names no hardware. Every option is accepted by preflight and create. */
+export interface ManagedTrainingOption {
+    /** `bf16` or `fp8`. */
+    precision: string;
+    /** `adamw_torch_fused` (16-bit state) or `adamw_8bit` (8-bit state). */
+    optim: string;
+    max_length: number;
+    /** Context lengths offered for this option: 2048 ... 131072. */
+    context_lengths: number[];
+}
+export interface ManagedTrainingOptions {
+    model_id: string;
+    /** Whether managed training is on in this environment. */
+    enabled: boolean;
+    /** Default (bf16, 16-bit state) first. Empty: not trainable yet. */
+    options: ManagedTrainingOption[];
+    all_context_bands: number[];
+    default_max_length: number;
+}
+export interface ManagedTrainingOptionsParams {
+    model: string;
+    /** sft (default), pt or cpt. */
+    method?: string;
+    /** lora (default), qlora or full. */
+    adapter?: string;
+}
+export interface TrainingCostEstimate {
+    train_rows: number;
+    mean_tokens_per_row: number;
+    max_length: number;
+    epochs: number;
+    max_steps?: number;
+    tokens_per_epoch: number;
+    estimated_tokens: number;
+    cents_per_million_tokens: number;
+    minimum_charge_cents: number;
+    estimated_cents: number;
+    image_tokens_included: boolean;
+    basis: string;
 }
 /** Parameters for `training.recommend()` (GET /api/training/recommend). */
 export interface TrainingRecommendParams {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "runbios-sdk",
-  "version": "0.5.0",
+  "version": "0.6.0-rc.369",
   "description": "Official TypeScript SDK for the Run BiOS training and deployment platform API",
   "type": "module",
   "main": "./dist/index.js",