@superlinked/sie-sdk 0.3.3 → 0.3.4

package/dist/index.d.cts

@@ -417,6 +417,14 @@ interface ExtractOptions {
     gpu?: string;
     /** Whether to wait for capacity */
     waitForCapacity?: boolean;
+    /**
+     * Adapter-specific runtime options forwarded to the server as
+     * `params.options`. Used for adapter knobs that aren't part of the
+     * core extract API — e.g. `{ overflow_policy: "error" }` for
+     * gliclass token-budget control. Mirrors the Python SDK's `options`
+     * keyword argument.
+     */
+    adapterOptions?: Record<string, unknown>;
 }
 /**
  * Helper to convert typed arrays to regular number array.
@@ -732,7 +740,7 @@ declare class SIEClient {
     private detectEndpointType;
 }
 
-declare const SDK_VERSION = "0.3.3";
+declare const SDK_VERSION = "0.3.4";
 
 /**
  * Helpers for converting SIE encode results to plain JavaScript types.
@@ -981,6 +989,26 @@ declare class ModelLoadFailedError extends ServerError {
         attempts?: number;
     });
 }
+/**
+ * Error when the request input exceeds the model's maximum token capacity.
+ *
+ * Thrown when the server returns HTTP `400 INPUT_TOO_LONG` for an
+ * extraction request. Distinct from generic {@link RequestError} so
+ * callers can branch on token-budget failures specifically (truncate
+ * the input client-side, switch to a longer-context model, or surface
+ * a targeted error to the end user) without parsing the error code.
+ *
+ * Subclass of {@link RequestError} so existing 4xx handlers continue
+ * to work; new code can catch {@link InputTooLongError} for tailored
+ * handling.
+ */
+declare class InputTooLongError extends RequestError {
+    /** The model that was requested */
+    readonly model: string | undefined;
+    constructor(message: string, options?: {
+        model?: string;
+    });
+}
 
 /**
  * MessagePack serialization with msgpack-numpy compatibility.
@@ -1086,4 +1114,4 @@ declare function toImageWireFormat(input: ImageInput, format?: "jpeg" | "png" |
  */
 declare function detectImageFormat(bytes: Uint8Array): "jpeg" | "png" | "webp" | "unknown";
 
-export { type CapacityInfo, type Classification, type ClusterStatusMessage, type ClusterSummary, type ClusterWorkerInfo, type DType, type DetectedObject, type EncodeOptions, type EncodeResult, type Entity, type ExtractOptions, type ExtractResult, type GPUMetrics, type ImageInput, type ImageWireFormat, type Item, LoraLoadingError, type ModelConfig, type ModelDims, type ModelInfo, ModelLoadFailedError, ModelLoadingError, type ModelState, type ModelStatus, type ModelSummary, type OutputType, PoolError, type PoolInfo, type PoolSpec, type PoolStatus, ProvisioningError, type Relation, RequestError, SDK_VERSION, SIEClient, type SIEClientOptions, SIEConnectionError, SIEError, type ScoreEntry, type ScoreOptions, type ScoreResult, ServerError, type ServerInfo, type SparseResult, type SparseVector, type StatusMessage, type TimingInfo, type WorkerInfo, type WorkerStatusMessage, denseEmbedding, detectImageFormat, multivectorEmbedding, normalizeSparseVector, packMessage, sparseEmbedding, sparseEmbeddingMap, toFloat32Array, toImageBytes, toImageWireFormat, toNumberArray, unpackMessage };
+export { type CapacityInfo, type Classification, type ClusterStatusMessage, type ClusterSummary, type ClusterWorkerInfo, type DType, type DetectedObject, type EncodeOptions, type EncodeResult, type Entity, type ExtractOptions, type ExtractResult, type GPUMetrics, type ImageInput, type ImageWireFormat, InputTooLongError, type Item, LoraLoadingError, type ModelConfig, type ModelDims, type ModelInfo, ModelLoadFailedError, ModelLoadingError, type ModelState, type ModelStatus, type ModelSummary, type OutputType, PoolError, type PoolInfo, type PoolSpec, type PoolStatus, ProvisioningError, type Relation, RequestError, SDK_VERSION, SIEClient, type SIEClientOptions, SIEConnectionError, SIEError, type ScoreEntry, type ScoreOptions, type ScoreResult, ServerError, type ServerInfo, type SparseResult, type SparseVector, type StatusMessage, type TimingInfo, type WorkerInfo, type WorkerStatusMessage, denseEmbedding, detectImageFormat, multivectorEmbedding, normalizeSparseVector, packMessage, sparseEmbedding, sparseEmbeddingMap, toFloat32Array, toImageBytes, toImageWireFormat, toNumberArray, unpackMessage };

package/dist/index.d.ts

@@ -417,6 +417,14 @@ interface ExtractOptions {
     gpu?: string;
     /** Whether to wait for capacity */
     waitForCapacity?: boolean;
+    /**
+     * Adapter-specific runtime options forwarded to the server as
+     * `params.options`. Used for adapter knobs that aren't part of the
+     * core extract API — e.g. `{ overflow_policy: "error" }` for
+     * gliclass token-budget control. Mirrors the Python SDK's `options`
+     * keyword argument.
+     */
+    adapterOptions?: Record<string, unknown>;
 }
 /**
  * Helper to convert typed arrays to regular number array.
@@ -732,7 +740,7 @@ declare class SIEClient {
     private detectEndpointType;
 }
 
-declare const SDK_VERSION = "0.3.3";
+declare const SDK_VERSION = "0.3.4";
 
 /**
  * Helpers for converting SIE encode results to plain JavaScript types.
@@ -981,6 +989,26 @@ declare class ModelLoadFailedError extends ServerError {
         attempts?: number;
     });
 }
+/**
+ * Error when the request input exceeds the model's maximum token capacity.
+ *
+ * Thrown when the server returns HTTP `400 INPUT_TOO_LONG` for an
+ * extraction request. Distinct from generic {@link RequestError} so
+ * callers can branch on token-budget failures specifically (truncate
+ * the input client-side, switch to a longer-context model, or surface
+ * a targeted error to the end user) without parsing the error code.
+ *
+ * Subclass of {@link RequestError} so existing 4xx handlers continue
+ * to work; new code can catch {@link InputTooLongError} for tailored
+ * handling.
+ */
+declare class InputTooLongError extends RequestError {
+    /** The model that was requested */
+    readonly model: string | undefined;
+    constructor(message: string, options?: {
+        model?: string;
+    });
+}
 
 /**
  * MessagePack serialization with msgpack-numpy compatibility.
@@ -1086,4 +1114,4 @@ declare function toImageWireFormat(input: ImageInput, format?: "jpeg" | "png" |
  */
 declare function detectImageFormat(bytes: Uint8Array): "jpeg" | "png" | "webp" | "unknown";
 
-export { type CapacityInfo, type Classification, type ClusterStatusMessage, type ClusterSummary, type ClusterWorkerInfo, type DType, type DetectedObject, type EncodeOptions, type EncodeResult, type Entity, type ExtractOptions, type ExtractResult, type GPUMetrics, type ImageInput, type ImageWireFormat, type Item, LoraLoadingError, type ModelConfig, type ModelDims, type ModelInfo, ModelLoadFailedError, ModelLoadingError, type ModelState, type ModelStatus, type ModelSummary, type OutputType, PoolError, type PoolInfo, type PoolSpec, type PoolStatus, ProvisioningError, type Relation, RequestError, SDK_VERSION, SIEClient, type SIEClientOptions, SIEConnectionError, SIEError, type ScoreEntry, type ScoreOptions, type ScoreResult, ServerError, type ServerInfo, type SparseResult, type SparseVector, type StatusMessage, type TimingInfo, type WorkerInfo, type WorkerStatusMessage, denseEmbedding, detectImageFormat, multivectorEmbedding, normalizeSparseVector, packMessage, sparseEmbedding, sparseEmbeddingMap, toFloat32Array, toImageBytes, toImageWireFormat, toNumberArray, unpackMessage };
+export { type CapacityInfo, type Classification, type ClusterStatusMessage, type ClusterSummary, type ClusterWorkerInfo, type DType, type DetectedObject, type EncodeOptions, type EncodeResult, type Entity, type ExtractOptions, type ExtractResult, type GPUMetrics, type ImageInput, type ImageWireFormat, InputTooLongError, type Item, LoraLoadingError, type ModelConfig, type ModelDims, type ModelInfo, ModelLoadFailedError, ModelLoadingError, type ModelState, type ModelStatus, type ModelSummary, type OutputType, PoolError, type PoolInfo, type PoolSpec, type PoolStatus, ProvisioningError, type Relation, RequestError, SDK_VERSION, SIEClient, type SIEClientOptions, SIEConnectionError, SIEError, type ScoreEntry, type ScoreOptions, type ScoreResult, ServerError, type ServerInfo, type SparseResult, type SparseVector, type StatusMessage, type TimingInfo, type WorkerInfo, type WorkerStatusMessage, denseEmbedding, detectImageFormat, multivectorEmbedding, normalizeSparseVector, packMessage, sparseEmbedding, sparseEmbeddingMap, toFloat32Array, toImageBytes, toImageWireFormat, toNumberArray, unpackMessage };

package/dist/index.js

@@ -107,6 +107,15 @@ var ModelLoadFailedError = class extends ServerError {
     this.attempts = options?.attempts ?? 1;
   }
 };
+var InputTooLongError = class extends RequestError {
+  /** The model that was requested */
+  model;
+  constructor(message, options) {
+    super(message, "INPUT_TOO_LONG", 400);
+    this.name = "InputTooLongError";
+    this.model = options?.model;
+  }
+};
 
 // src/internal/constants.ts
 var MSGPACK_CONTENT_TYPE = "application/msgpack";
@@ -378,6 +387,14 @@ async function throwIfModelLoadFailed(response, model) {
     attempts
   });
 }
+async function throwIfInputTooLong(response, model) {
+  if (response.status !== 400) return;
+  const detail = await getErrorDetail(response.clone());
+  if (!detail) return;
+  if (detail.code !== "INPUT_TOO_LONG") return;
+  const message = typeof detail.message === "string" ? detail.message : "Input exceeds the model's maximum token capacity";
+  throw new InputTooLongError(message, { model });
+}
 async function handleError(response, gpu) {
   const { status } = response;
   const detail = await getErrorDetail(response.clone());
@@ -411,6 +428,9 @@ async function handleError(response, gpu) {
     throw new ProvisioningError(message, gpu, retryAfter);
   }
   if (status >= HTTP_CLIENT_ERROR_MIN && status <= HTTP_CLIENT_ERROR_MAX) {
+    if (status === 400 && code === "INPUT_TOO_LONG") {
+      throw new InputTooLongError(message);
+    }
     throw new RequestError(message, code, status);
   }
   if (status >= HTTP_SERVER_ERROR_MIN && status <= HTTP_SERVER_ERROR_MAX) {
@@ -529,7 +549,7 @@ function parseCapacityInfo(data, gpuFilter) {
 }
 
 // src/version.ts
-var SDK_VERSION = "0.3.3";
+var SDK_VERSION = "0.3.4";
 
 // src/client.ts
 function sleep(ms) {
@@ -818,6 +838,9 @@ var SIEClient = class {
     if (options.threshold !== void 0) {
       params.threshold = options.threshold;
     }
+    if (options.adapterOptions !== void 0) {
+      params.options = options.adapterOptions;
+    }
     body.params = params;
     const waitForCapacity = options.waitForCapacity ?? this.defaultWaitForCapacity;
     const { pool, gpu } = this.parseGpuParam(options.gpu);
@@ -1234,6 +1257,7 @@ var SIEClient = class {
         continue;
       }
       await throwIfModelLoadFailed(response, model);
+      await throwIfInputTooLong(response, model);
       if (response.status === 503) {
         const clonedResponse = response.clone();
         const errorCode = await getErrorCode(clonedResponse);
@@ -1555,6 +1579,6 @@ function detectImageFormat(bytes) {
   return "unknown";
 }
 
-export { LoraLoadingError, ModelLoadFailedError, ModelLoadingError, PoolError, ProvisioningError, RequestError, SDK_VERSION, SIEClient, SIEConnectionError, SIEError, ServerError, denseEmbedding, detectImageFormat, maxsim, maxsimBatch, maxsimDocuments, multivectorEmbedding, normalizeSparseVector, packMessage, sparseEmbedding, sparseEmbeddingMap, toFloat32Array, toImageBytes, toImageWireFormat, toNumberArray, unpackMessage };
+export { InputTooLongError, LoraLoadingError, ModelLoadFailedError, ModelLoadingError, PoolError, ProvisioningError, RequestError, SDK_VERSION, SIEClient, SIEConnectionError, SIEError, ServerError, denseEmbedding, detectImageFormat, maxsim, maxsimBatch, maxsimDocuments, multivectorEmbedding, normalizeSparseVector, packMessage, sparseEmbedding, sparseEmbeddingMap, toFloat32Array, toImageBytes, toImageWireFormat, toNumberArray, unpackMessage };
 //# sourceMappingURL=index.js.map
 //# sourceMappingURL=index.js.map