runbios-sdk 0.2.19-dev.277 → 0.2.19-dev.281

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 = "0.2.19-dev.277";
+export declare const VERSION = "0.2.19-dev.281";
 export declare class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     readonly models: Models;
@@ -75,7 +75,7 @@ export { Training } from './resources/training.js';
 export { Wallet } from './resources/wallet.js';
 export { GPU, type GPURecommendation } from './resources/gpu.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 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, TrainingListParams, TrainingListResponse, TrainingJob, TrainingMetrics, TrainingResourceMetrics, TrainingDeviceMetrics, MetricPoint, MetricGraphConfig, TrainingCheckpoint, TrainingLogs, TrainingLogEntry, TrainingStopResponse, TrainingResumeResponse, CanonicalTrainingRequest, TrainingPreflightDataset, TrainingPreflightWarning, TrainingPreflightResponse, TrainingCapabilityChoice, TrainingConfigFieldCapability, TrainingCapabilities, 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, NotifyState, EvaluationStatus, EvaluationItemStatus, EvaluationWinner, EvaluationItemWinnerFilter, EvaluationWarningCode, TrainingToolCall, TrainingMessage, TrainingRuleServing, TrainingRunPromoteRequest, TrainingRunRejectRequest, TrainingRunRollbackRequest, TrainingRunCancelRequest, TrainingRunListParams, TrainingRun, TrainingRunSummary, TrainingRunEvent, TrainingRunLinks, TrainingRunActions, EvaluationRef, EvaluationDecoding, EvaluationJudgeDimension, EvaluationDimensionScore, EvaluationWarning, EvaluationMargin, Evaluation, EvaluationItemSide, EvaluationItem, EvaluationItemListParams, BenchmarkStatus, BenchmarkSourceKind, BenchmarkRunStatus, BenchmarkScoring, BenchmarkDimensionScore, Benchmark, BenchmarkItem, BenchmarkRun, BenchmarkHistoryPoint, BenchmarkSource, BenchmarkCreateParams, BenchmarkListParams, BenchmarkItemListParams, BenchmarkHistoryParams, BenchmarkRetireParams, InferenceAlias, InferenceAliasRequest, TrainingRuleDeleteResponse, TrainingRunListResponse, TrainingRunResponse, TrainingRunActionResponse, EvaluationItemsResponse, InferenceAliasListResponse, InferenceAliasResponse, InferenceAliasDeleteResponse, BenchmarkListResponse, BenchmarkItemsResponse, BenchmarkHistoryResponse, TrainingRuleRunRefusalCode, TrainingRuleSaveRefusalCode, TrainingRuleMonthlyLimitRefusal, PipelineStatus, PipelineActiveRun, PipelineVersion, PipelineAttempt, PipelineAttemptTrigger, PipelineAttemptOutcome, Pipeline, PipelineListParams, PipelineListResponse, PipelineResponse, PipelineTrainWhen, PipelineSchedule, PipelineWriteRequest, PipelineMutationResponse, PipelineImportParams, PipelineFloors, LoopModel, LoopModelsResponse, LoopTracePipeline, LoopTracePipelinesResponse, LoopMetricsHistoryParams, LoopMetricPoint, LoopSignalDay, LoopMetricsHistory, PromotionPolicyState, RuleBenchmark, PipelineBenchmarksResponse, PipelineBenchmarkAttachParams, PipelineBenchmarkMoveParams, CompareBenchmark, VersionBenchmarkScore, PromotionMeasurement, PromotionOutcome, VersionScores, VersionScoreDifference, VersionCompareResponse, VersionCompareParams, } 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, TrainingListParams, TrainingListResponse, TrainingJob, TrainingMetrics, TrainingResourceMetrics, TrainingDeviceMetrics, MetricPoint, MetricGraphConfig, TrainingCheckpoint, TrainingLogs, TrainingLogEntry, TrainingStopResponse, TrainingResumeResponse, CanonicalTrainingRequest, TrainingPreflightDataset, TrainingPreflightWarning, TrainingPreflightResponse, TrainingCapabilityChoice, TrainingConfigFieldCapability, TrainingCapabilities, 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, EvaluationWinner, EvaluationItemWinnerFilter, EvaluationWarningCode, TrainingToolCall, TrainingMessage, TrainingRuleServing, TrainingRunPromoteRequest, TrainingRunRejectRequest, TrainingRunRollbackRequest, TrainingRunCancelRequest, TrainingRunListParams, TrainingRun, TrainingRunSummary, TrainingRunEvent, TrainingRunLinks, TrainingRunActions, EvaluationRef, EvaluationDecoding, EvaluationJudgeDimension, EvaluationDimensionScore, EvaluationWarning, EvaluationMargin, Evaluation, EvaluationItemSide, EvaluationItem, EvaluationItemListParams, BenchmarkStatus, BenchmarkSourceKind, BenchmarkRunStatus, BenchmarkScoring, BenchmarkDimensionScore, Benchmark, BenchmarkItem, BenchmarkRun, BenchmarkHistoryPoint, BenchmarkSource, BenchmarkCreateParams, BenchmarkListParams, BenchmarkItemListParams, BenchmarkHistoryParams, BenchmarkRetireParams, InferenceAlias, InferenceAliasRequest, TrainingRuleDeleteResponse, TrainingRunListResponse, TrainingRunResponse, TrainingRunActionResponse, EvaluationItemsResponse, InferenceAliasListResponse, InferenceAliasResponse, InferenceAliasDeleteResponse, BenchmarkListResponse, BenchmarkItemsResponse, BenchmarkHistoryResponse, TrainingRuleRunRefusalCode, TrainingRuleSaveRefusalCode, TrainingRuleMonthlyLimitRefusal, PipelineStatus, PipelineActiveRun, PipelineVersion, PipelineAttempt, PipelineAttemptTrigger, PipelineAttemptOutcome, Pipeline, PipelineListParams, PipelineListResponse, PipelineResponse, PipelineTrainWhen, PipelineSchedule, PipelineWriteRequest, PipelineMutationResponse, PipelineImportParams, PipelineFloors, LoopModel, LoopModelsResponse, LoopTracePipeline, LoopTracePipelinesResponse, LoopMetricsHistoryParams, LoopMetricPoint, LoopSignalDay, LoopMetricsHistory, PromotionPolicyState, RuleBenchmark, PipelineBenchmarksResponse, PipelineBenchmarkAttachParams, PipelineBenchmarkMoveParams, CompareBenchmark, VersionBenchmarkScore, PromotionMeasurement, PromotionOutcome, VersionScores, VersionScoreDifference, VersionCompareResponse, VersionCompareParams, } 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 = '0.2.19-dev.277';
+export const VERSION = '0.2.19-dev.281';
 export class RunBiOS {
     /** Search models, fetch configs, check adapter compatibility. */
     models;

package/dist/resources/loop.d.ts

@@ -402,7 +402,9 @@ export declare class Loop {
      * numbers `minimums` shows. 503 `AGENT_COULD_NOT_START` means the key the
      * attempt would be scored with could not be renewed just then: nothing was
      * started, so press train now again in a minute. Every refusal is a
-     * `TrainingRuleRunRefusalCode`.
+     * `TrainingRuleRunRefusalCode`. When the attempt in flight is waiting for
+     * GPUs to compare (`active_run.state` `waiting_gpus`), this books its two
+     * comparison machines again at once instead of starting another attempt.
      */
     runPipeline(id: string): Promise<PipelineMutationResponse>;
     /**

package/dist/resources/loop.js

@@ -555,7 +555,9 @@ export class Loop {
      * numbers `minimums` shows. 503 `AGENT_COULD_NOT_START` means the key the
      * attempt would be scored with could not be renewed just then: nothing was
      * started, so press train now again in a minute. Every refusal is a
-     * `TrainingRuleRunRefusalCode`.
+     * `TrainingRuleRunRefusalCode`. When the attempt in flight is waiting for
+     * GPUs to compare (`active_run.state` `waiting_gpus`), this books its two
+     * comparison machines again at once instead of starting another attempt.
      */
     async runPipeline(id) {
         return this._http.fetchPost(`/api/loop/pipelines/${encodeURIComponent(id)}/run`, {});

package/dist/types.d.ts

@@ -2611,6 +2611,11 @@ export interface LoopTrace {
     verdict?: LoopVerdict;
     /** Who gave that verdict. */
     verdict_source?: LoopSource;
+    /**
+     * The score a `scored` verdict carries: above zero it trains the answer as
+     * it stands (good feedback), zero or below it trains nothing (bad).
+     */
+    verdict_score?: number;
     /** Its tags. On the single read only. */
     labels?: LoopLabel[];
     /**
@@ -2861,12 +2866,22 @@ export interface TrainingRecipe {
     lora_rank: number | null;
     lora_alpha: number | null;
 }
-export type TrainingRunState = 'built' | 'registering' | 'dataset_validating' | 'preflighting' | 'job_creating' | 'training' | 'checkpoint_ready' | 'candidate_booking' | 'candidate_running' | 'evaluating' | 'reported' | 'awaiting_review' | 'promoting' | 'waiting_funds' | 'promoted' | 'rejected' | 'expired' | 'failed' | 'budget_stopped' | 'cancelled' | 'superseded' | 'rolled_back';
+export type TrainingRunState = 'built' | 'registering' | 'dataset_validating' | 'preflighting' | 'job_creating' | 'training' | 'checkpoint_ready' | 'candidate_booking' | 'candidate_running' | 'evaluating' | 'reported' | 'awaiting_review' | 'promoting' | 'waiting_funds'
+/**
+ * Trained, and waiting for GPUs to run the new model and the one it is
+ * compared with side by side: one machine got a GPU and the other did not
+ * within 20 minutes, so both were stopped. Each try can bill the machine
+ * that got a GPU for up to those 20 minutes; nothing bills between tries.
+ * It keeps what it trained, books both again after 30 minutes, then 1, 2
+ * and 4 hours, and ends (`failed`, `COMPARISON_NO_GPUS`) after four tries or
+ * 24 hours (`COMPARISON_WORKSPACE_FULL` when what was missing was room in
+ * the workspace). Train now books them again at once.
+ */
+ | 'waiting_gpus' | 'promoted' | 'rejected' | 'expired' | 'failed' | 'budget_stopped' | 'cancelled' | 'superseded' | 'rolled_back';
 /** The closed set a run never leaves. */
 export declare const TERMINAL_RUN_STATES: readonly TrainingRunState[];
 export type TrainingVerdict = 'better' | 'not_better' | 'inconclusive' | 'not_evaluated';
 export type TrainingDecision = 'auto_promoted' | 'promoted' | 'rejected' | 'auto_rejected' | 'rolled_back';
-export type NotifyState = 'none' | 'pending' | 'sending' | 'sent' | 'suppressed' | 'failed';
 export type EvaluationStatus = 'open' | 'done' | 'failed' | 'budget_stopped';
 export type EvaluationItemStatus = 'pending' | 'working' | 'done' | 'failed';
 export type EvaluationWinner = 'candidate' | 'incumbent' | 'tie';
@@ -2953,7 +2968,19 @@ export interface TrainingRun {
      */
     error_next_step: string | null;
     billed_training_cents: number;
+    /**
+     * The comparison machines: each from the first time it was on a GPU to its
+     * stop, less any time back in the GPU queue, at the price the platform
+     * placed it at.
+     */
     billed_candidate_cents: number;
+    /**
+     * True when a comparison machine was charged: the figure is then the most
+     * it can have cost ("up to"), since it counts each machine from its first
+     * boot where the platform bills from its first answer (and at the hourly
+     * cap when no placement price was reported).
+     */
+    billed_candidate_up_to: boolean;
     spent_eval_cents: number;
     train_rows: number | null;
     holdout_rows: number | null;
@@ -3008,10 +3035,6 @@ export interface TrainingRun {
     decided_at: string | null;
     auto_promote_at: string | null;
     review_deadline_at: string | null;
-    notify_state: NotifyState;
-    notify_event: string | null;
-    notify_error: string | null;
-    notify_attempts: number;
     created_at: string;
     updated_at: string;
     finished_at: string | null;
@@ -3049,6 +3072,8 @@ export interface TrainingRunSummary {
     holdout_rows: number | null;
     billed_training_cents: number;
     billed_candidate_cents: number;
+    /** See {@link TrainingRun.billed_candidate_up_to}. */
+    billed_candidate_up_to: boolean;
     spent_eval_cents: number;
     /** The fourth money column. A row that leaves it out adds up short. */
     benchmark_spent_cents: number;
@@ -3057,7 +3082,6 @@ export interface TrainingRunSummary {
     evaluation_id: string | null;
     review_deadline_at: string | null;
     auto_promote_at: string | null;
-    notify_state: NotifyState;
     created_at: string;
     finished_at: string | null;
 }
@@ -3515,8 +3539,8 @@ export type TrainingRuleSaveRefusalCode = 'PIPELINE_NAME_TAKEN' | 'OWN_TAG_TAKEN
  *
  * The test is the worst case, not the average: a run may start only if the
  * most the rule's runs this UTC calendar month can have cost
- * (`month_spent_cents`), plus the most the run about to start may spend (its
- * three ceilings added up), fits under the limit.
+ * (`month_spent_cents`), plus the most the run about to start may spend
+ * (`run_max_cents`), fits under the limit.
  */
 export interface TrainingRuleMonthlyLimitRefusal {
     error: {
@@ -3527,7 +3551,11 @@ export interface TrainingRuleMonthlyLimitRefusal {
     month_spent_cents: number;
     /** The rule's monthly limit. */
     monthly_ceiling_cents: number;
-    /** The most one run of this rule may spend: training + candidate + evaluation ceilings. */
+    /**
+     * The most one attempt of this rule may spend: its training, its comparison
+     * machine and its judging, and -- on a pipeline, while no version is live --
+     * the base model's own comparison machine beside the new model's.
+     */
     run_max_cents: number;
     /**
      * The first instant of the next UTC month, when the counted spend starts
@@ -3571,6 +3599,11 @@ export interface PipelineActiveRun {
     gpu_seconds: number | null;
     /** What it has been billed so far, in cents. */
     cost_cents: number;
+    /**
+     * True when a comparison machine was charged in `cost_cents`: the figure is
+     * the most it can have cost, so show it as "up to".
+     */
+    cost_up_to: boolean;
     /** The version number it will take if it is put live. */
     will_be_version: number;
     /** Always null: nothing in flight is a version yet. */
@@ -3643,6 +3676,14 @@ export interface PipelineVersion {
      * not what it did.
      */
     cost_includes_candidate_ceiling: boolean;
+    /**
+     * True when `cost_cents` is the most the version can have cost rather than
+     * what it did: its comparison machines at their ceilings
+     * (`cost_includes_candidate_ceiling`), or one charged (counted from its
+     * first boot, where the platform bills from its first answer). Show it as
+     * "up to".
+     */
+    cost_up_to: boolean;
     /** Machine time the attempt held (training and comparison); null when unknown. */
     gpu_seconds: number | null;
     created_at: string;
@@ -3691,6 +3732,8 @@ export interface PipelineAttempt {
     gpu_seconds: number | null;
     /** As the versions are priced once it has ended; as recorded so far while it runs. */
     cost_cents: number;
+    /** See {@link PipelineVersion.cost_up_to}. */
+    cost_up_to: boolean;
     verdict: TrainingVerdict | null;
     win_rate: number | null;
     evaluation_id: string | null;
@@ -3820,10 +3863,11 @@ export interface Pipeline {
      * this UTC calendar month -- the figure the limit is enforced against. An
      * upper bound, not an exact spend. A run counts toward the month it was
      * created in, and a run still in progress also counts toward the current
-     * month, at its three ceilings or what it has been billed when that is
-     * more. A finished run counts what it was billed, and its comparison
-     * machine is billed at the most it could have cost (the hourly cap for the
-     * time it was up, never more than the candidate ceiling). A run created in
+     * month, at the most it may cost (as `run_max_cents` counts it, the base
+     * model's comparison machine included) or what it has been billed when that
+     * is more. A finished run counts what it was billed, and its comparison
+     * machines are billed at the most they could have cost (each machine's
+     * hourly cap for the time it was up, never more than its own amount). A run created in
      * an earlier month that finishes in this one counts here only while it is
      * still running.
      */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "runbios-sdk",
-  "version": "0.2.19-dev.277",
+  "version": "0.2.19-dev.281",
   "description": "Official TypeScript SDK for the Run BiOS training and deployment platform API",
   "type": "module",
   "main": "./dist/index.js",