@use-lattice/litmus 0.121.3

package/dist/src/sagemaker-Cno2V-Sx.js

@@ -0,0 +1,689 @@
+import { b as getEnvInt, r as logger, x as getEnvString, y as getEnvFloat } from "./logger-C40ZGil9.js";
+import { n as transform } from "./transform-yWaShiKr.js";
+import { t as telemetry } from "./telemetry-Bg2WqF79.js";
+import { z } from "zod";
+import crypto from "crypto";
+//#region src/providers/sagemaker.ts
+/**
+* Sleep utility function for implementing delays
+* @param ms Milliseconds to sleep
+* @returns Promise that resolves after the specified delay
+*/
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+const SUPPORTED_MODEL_TYPES = [
+	"openai",
+	"llama",
+	"huggingface",
+	"jumpstart",
+	"custom"
+];
+/**
+* Zod schema for validating SageMaker options
+*/
+const SageMakerConfigSchema = z.strictObject({
+	accessKeyId: z.string().optional(),
+	profile: z.string().optional(),
+	region: z.string().optional(),
+	secretAccessKey: z.string().optional(),
+	sessionToken: z.string().optional(),
+	endpoint: z.string().optional(),
+	contentType: z.string().optional(),
+	acceptType: z.string().optional(),
+	maxTokens: z.number().optional(),
+	temperature: z.number().optional(),
+	topP: z.number().optional(),
+	stopSequences: z.array(z.string()).optional(),
+	delay: z.number().optional(),
+	transform: z.string().optional(),
+	modelType: z.enum(SUPPORTED_MODEL_TYPES).optional(),
+	responseFormat: z.strictObject({
+		type: z.string().optional(),
+		path: z.string().optional()
+	}).optional(),
+	basePath: z.string().optional()
+});
+/**
+* Base class for SageMaker providers with common functionality
+*/
+var SageMakerGenericProvider = class {
+	env;
+	sagemakerRuntime;
+	config;
+	endpointName;
+	delay;
+	transform;
+	providerId;
+	constructor(endpointName, options) {
+		const { config, id, env, delay, transform } = options;
+		this.env = env;
+		this.endpointName = endpointName;
+		try {
+			SageMakerConfigSchema.parse(config);
+		} catch (error) {
+			logger.warn(`Error validating SageMaker config\nConfig: ${JSON.stringify(config)}\n${error instanceof z.ZodError ? z.prettifyError(error) : error}`);
+		}
+		this.config = config ?? {};
+		this.delay = delay || this.config.delay;
+		this.transform = transform || this.config.transform;
+		this.providerId = id;
+		telemetry.record("feature_used", { feature: "sagemaker" });
+	}
+	id() {
+		return this.providerId || `sagemaker:${this.endpointName}`;
+	}
+	toString() {
+		return `[Amazon SageMaker Provider ${this.endpointName}]`;
+	}
+	/**
+	* Get AWS credentials from config or environment
+	*/
+	async getCredentials() {
+		if (this.config.accessKeyId && this.config.secretAccessKey) {
+			logger.debug("Using explicit credentials from config");
+			return {
+				accessKeyId: this.config.accessKeyId,
+				secretAccessKey: this.config.secretAccessKey,
+				sessionToken: this.config.sessionToken
+			};
+		}
+		if (this.config.profile) {
+			logger.debug(`Using AWS profile: ${this.config.profile}`);
+			try {
+				const { fromSSO } = await import("@aws-sdk/credential-provider-sso");
+				return fromSSO({ profile: this.config.profile });
+			} catch {
+				throw new Error(`Failed to load AWS SSO profile. Please install @aws-sdk/credential-provider-sso`);
+			}
+		}
+		logger.debug("Using default AWS credentials from environment");
+	}
+	/**
+	* Initialize and return the SageMaker runtime client
+	*/
+	async getSageMakerRuntimeInstance() {
+		if (!this.sagemakerRuntime) try {
+			const { SageMakerRuntimeClient } = await import("@aws-sdk/client-sagemaker-runtime");
+			const credentials = await this.getCredentials();
+			this.sagemakerRuntime = new SageMakerRuntimeClient({
+				region: this.getRegion(),
+				maxAttempts: getEnvInt("AWS_SAGEMAKER_MAX_RETRIES", 3),
+				retryMode: "adaptive",
+				...credentials ? { credentials } : {}
+			});
+			logger.debug(`SageMaker client initialized for region ${this.getRegion()}`);
+		} catch {
+			throw new Error("The @aws-sdk/client-sagemaker-runtime package is required. Please install it with: npm install @aws-sdk/client-sagemaker-runtime");
+		}
+		return this.sagemakerRuntime;
+	}
+	/**
+	* Get AWS region from config or environment
+	*/
+	getRegion() {
+		return this.config?.region || this.env?.AWS_REGION || getEnvString("AWS_REGION") || getEnvString("AWS_DEFAULT_REGION") || "us-east-1";
+	}
+	/**
+	* Get SageMaker endpoint name
+	*/
+	getEndpointName() {
+		return this.config?.endpoint || this.endpointName;
+	}
+	/**
+	* Get content type for request
+	*/
+	getContentType() {
+		return this.config?.contentType || "application/json";
+	}
+	/**
+	* Get accept type for response
+	*/
+	getAcceptType() {
+		return this.config?.acceptType || "application/json";
+	}
+	/**
+	* Apply transformation to a prompt if a transform function is specified
+	* @param prompt The original prompt to transform
+	* @param context Optional context information for the transformation
+	* @returns The transformed prompt, or the original if no transformation is applied
+	*/
+	async applyTransformation(prompt, context) {
+		if (!this.transform) return prompt;
+		try {
+			const transformContext = {
+				vars: context?.vars || {},
+				prompt: context?.prompt || { raw: prompt },
+				uuid: `sagemaker-${this.endpointName}-${Date.now()}`
+			};
+			const transformFn = this.transform || context?.originalProvider?.transform;
+			if (!transformFn) return prompt;
+			logger.debug(`Applying transform to prompt for SageMaker endpoint ${this.getEndpointName()}`);
+			if (typeof transformFn === "string" && !transformFn.startsWith("file://")) try {
+				if (transformFn.includes("=>")) {
+					const result = new Function("prompt", "context", `try { return (${transformFn})(prompt, context); } catch(e) { throw new Error("Transform function error: " + e.message); }`)(prompt, transformContext);
+					if (result === void 0 || result === null) {
+						logger.debug("Transform function returned null or undefined, using original prompt");
+						return prompt;
+					}
+					if (typeof result === "string") return result;
+					else if (typeof result === "object") return JSON.stringify(result);
+					else return String(result);
+				} else {
+					const result = new Function("prompt", "context", `try { ${transformFn} } catch(e) { throw new Error("Transform function error: " + e.message); }`)(prompt, transformContext);
+					if (result === void 0 || result === null) {
+						logger.debug("Transform function returned null or undefined, using original prompt");
+						return prompt;
+					}
+					if (typeof result === "string") return result;
+					else if (typeof result === "object") return JSON.stringify(result);
+					else return String(result);
+				}
+			} catch (transformError) {
+				logger.error(`Error executing inline transform: ${transformError}`);
+			}
+			else try {
+				const { TransformInputType } = await import("./transform-yWaShiKr.js").then((n) => n.r);
+				const transformed = await transform(transformFn, prompt, transformContext, false, TransformInputType.OUTPUT);
+				if (transformed === void 0 || transformed === null) {
+					logger.debug("Transform function returned null or undefined, using original prompt");
+					return prompt;
+				}
+				if (typeof transformed === "string") return transformed;
+				else if (typeof transformed === "object") return JSON.stringify(transformed);
+				else return String(transformed);
+			} catch (transformError) {
+				logger.error(`Error using transform utility: ${transformError}`);
+			}
+			logger.warn(`Transform did not produce a valid result, using original prompt`);
+			return prompt;
+		} catch (_) {
+			logger.error(`Error applying transform to prompt: ${_}`);
+			return prompt;
+		}
+	}
+	/**
+	* Extracts data from a response using a path expression
+	* Supports JavaScript expressions and file-based transforms
+	*/
+	async extractFromPath(responseJson, pathExpression) {
+		if (!pathExpression) return responseJson;
+		try {
+			if (pathExpression.startsWith("file://")) try {
+				const { TransformInputType } = await import("./transform-yWaShiKr.js").then((n) => n.r);
+				const transformedResult = await transform(pathExpression, responseJson, { prompt: {} }, false, TransformInputType.OUTPUT);
+				return transformedResult !== void 0 && transformedResult !== null ? transformedResult : responseJson;
+			} catch (error) {
+				logger.warn(`Failed to transform response using file: ${error}`);
+				return responseJson;
+			}
+			try {
+				const result = new Function("json", `try { return ${pathExpression}; } catch(e) { return undefined; }`)(responseJson);
+				if (result === void 0) {
+					logger.warn(`Path expression "${pathExpression}" did not match any data in the response`);
+					logger.debug(`Response JSON structure: ${JSON.stringify(responseJson).substring(0, 200)}...`);
+					return responseJson;
+				}
+				return result;
+			} catch (error) {
+				logger.warn(`Failed to evaluate expression "${pathExpression}": ${error}`);
+				return responseJson;
+			}
+		} catch (error) {
+			logger.warn(`Failed to extract data using path expression "${pathExpression}": ${error}`);
+			logger.debug(`Response JSON structure: ${JSON.stringify(responseJson).substring(0, 200)}...`);
+			return responseJson;
+		}
+	}
+};
+/**
+* Provider for text generation with SageMaker endpoints
+*/
+var SageMakerCompletionProvider = class extends SageMakerGenericProvider {
+	modelType;
+	constructor(endpointName, options) {
+		super(endpointName, options);
+		this.modelType = this.parseModelType(options.config?.modelType);
+	}
+	/**
+	* Model type must be specified within the id or the `config.modelType` field.
+	*/
+	parseModelType(modelType) {
+		const match = this.id().match(/^sagemaker:(?<modelType>.+):.+$/);
+		if (match) {
+			const modelTypeFromId = match.groups.modelType;
+			if (SUPPORTED_MODEL_TYPES.includes(modelTypeFromId)) return modelTypeFromId;
+			else throw new Error(`Invalid model type "${modelTypeFromId}" in provider ID. Valid types are: ${SUPPORTED_MODEL_TYPES.join(", ")}`);
+		}
+		if (modelType) if (SUPPORTED_MODEL_TYPES.includes(modelType)) return modelType;
+		else throw new Error(`Invalid model type "${modelType}" in \`config.modelType\`. Valid types are: ${SUPPORTED_MODEL_TYPES.join(", ")}`);
+		throw new Error("Model type must be set either in `config.modelType` or as part of the Provider ID, for example: \"sagemaker:<model_type>:<endpoint>\"");
+	}
+	/**
+	* Format the request payload based on model type
+	*/
+	formatPayload(prompt) {
+		const maxTokens = this.config.maxTokens ?? getEnvInt("AWS_SAGEMAKER_MAX_TOKENS") ?? 1024;
+		const temperature = typeof this.config.temperature === "number" ? this.config.temperature : getEnvFloat("AWS_SAGEMAKER_TEMPERATURE") ?? .7;
+		const topP = typeof this.config.topP === "number" ? this.config.topP : getEnvFloat("AWS_SAGEMAKER_TOP_P") ?? 1;
+		const stopSequences = this.config.stopSequences || [];
+		let payload;
+		logger.debug(`Formatting payload for model type: ${this.modelType}`);
+		switch (this.modelType) {
+			case "openai":
+				try {
+					const messages = JSON.parse(prompt);
+					if (Array.isArray(messages)) payload = {
+						messages,
+						max_tokens: maxTokens,
+						temperature,
+						top_p: topP,
+						stop: stopSequences.length > 0 ? stopSequences : void 0
+					};
+					else throw new Error("Not valid messages format");
+				} catch {
+					payload = {
+						prompt,
+						max_tokens: maxTokens,
+						temperature,
+						top_p: topP,
+						stop: stopSequences.length > 0 ? stopSequences : void 0
+					};
+				}
+				break;
+			case "llama":
+				try {
+					const messages = JSON.parse(prompt);
+					if (Array.isArray(messages)) payload = {
+						inputs: messages,
+						parameters: {
+							max_new_tokens: maxTokens,
+							temperature,
+							top_p: topP,
+							stop: stopSequences.length > 0 ? stopSequences : void 0
+						}
+					};
+					else throw new Error("Not valid messages format");
+				} catch {
+					payload = {
+						inputs: prompt,
+						parameters: {
+							max_new_tokens: maxTokens,
+							temperature,
+							top_p: topP,
+							stop: stopSequences.length > 0 ? stopSequences : void 0
+						}
+					};
+				}
+				break;
+			case "jumpstart":
+				payload = {
+					inputs: prompt,
+					parameters: {
+						max_new_tokens: maxTokens,
+						temperature,
+						top_p: topP,
+						do_sample: temperature > 0
+					}
+				};
+				break;
+			case "huggingface":
+				payload = {
+					inputs: prompt,
+					parameters: {
+						max_new_tokens: maxTokens,
+						temperature,
+						top_p: topP,
+						do_sample: temperature > 0,
+						return_full_text: false
+					}
+				};
+				break;
+			default:
+				try {
+					payload = JSON.parse(prompt);
+				} catch {
+					payload = { prompt };
+				}
+				break;
+		}
+		return JSON.stringify(payload);
+	}
+	/**
+	* Parse the response from SageMaker endpoint
+	*/
+	async parseResponse(responseBody) {
+		let responseJson;
+		logger.debug(`Parsing response for model type: ${this.modelType}`);
+		try {
+			responseJson = JSON.parse(responseBody);
+		} catch {
+			logger.debug("Response is not JSON, returning as-is");
+			return responseBody;
+		}
+		if (this.config.responseFormat?.path) try {
+			const pathExpression = this.config.responseFormat.path;
+			return await this.extractFromPath(responseJson, pathExpression);
+		} catch (error) {
+			logger.warn(`Failed to extract from path: ${this.config.responseFormat.path}, Error: ${error}`);
+			logger.debug(`Response JSON structure: ${JSON.stringify(responseJson).substring(0, 200)}...`);
+			return responseJson;
+		}
+		if (responseJson.generated_text) {
+			logger.debug("Detected JumpStart model response format with generated_text field");
+			return responseJson.generated_text;
+		}
+		switch (this.modelType) {
+			case "openai": return responseJson.choices?.[0]?.message?.content || responseJson.choices?.[0]?.text || responseJson.generation || responseJson;
+			case "llama": return responseJson.generation || responseJson.choices?.[0]?.message?.content || responseJson.choices?.[0]?.text || responseJson;
+			case "huggingface": return Array.isArray(responseJson) ? responseJson[0]?.generated_text || responseJson[0] : responseJson.generated_text || responseJson;
+			case "jumpstart": return responseJson.generated_text || responseJson;
+			default: return responseJson.output || responseJson.generation || responseJson.response || responseJson.text || responseJson.generated_text || responseJson.choices?.[0]?.message?.content || responseJson.choices?.[0]?.text || responseJson;
+		}
+	}
+	/**
+	* Generate a consistent cache key for SageMaker requests
+	* Uses crypto.createHash to generate a shorter, more efficient key
+	*/
+	getCacheKey(prompt) {
+		const configForKey = {
+			endpoint: this.getEndpointName(),
+			modelType: this.config.modelType,
+			contentType: this.getContentType(),
+			acceptType: this.getAcceptType(),
+			maxTokens: this.config.maxTokens,
+			temperature: this.config.temperature,
+			topP: this.config.topP,
+			region: this.getRegion()
+		};
+		const configStr = JSON.stringify(configForKey);
+		const promptHash = crypto.createHash("sha256").update(prompt).digest("hex").substring(0, 16);
+		const configHash = crypto.createHash("sha256").update(configStr).digest("hex").substring(0, 8);
+		return `sagemaker:v1:${this.getEndpointName()}:${promptHash}:${configHash}`;
+	}
+	/**
+	* Invoke SageMaker endpoint for text generation with caching, delay support, and transformations
+	*/
+	async callApi(prompt, context, _options) {
+		const { isCacheEnabled, getCache } = await import("./cache-BkrqU5Ba.js").then((n) => n.t);
+		const delayMs = context?.originalProvider?.delay || this.delay;
+		const transformedPrompt = await this.applyTransformation(prompt, context);
+		const isTransformed = transformedPrompt !== prompt;
+		if (isTransformed) {
+			logger.debug(`Prompt transformed for SageMaker endpoint ${this.getEndpointName()}`);
+			logger.debug(`Original: ${prompt.substring(0, 100)}${prompt.length > 100 ? "..." : ""}`);
+			logger.debug(`Transformed: ${transformedPrompt.substring(0, 100)}${transformedPrompt.length > 100 ? "..." : ""}`);
+		}
+		const bustCache = context?.bustCache ?? context?.debug === true;
+		if (isCacheEnabled() && !bustCache) {
+			const cacheKey = this.getCacheKey(transformedPrompt);
+			const cachedResult = await (getCache ? getCache() : await import("./cache-BkrqU5Ba.js").then((n) => n.t).then((m) => m.getCache())).get(cacheKey);
+			if (cachedResult) {
+				logger.debug(`Using cached SageMaker response for ${this.getEndpointName()}`);
+				try {
+					const parsedResult = JSON.parse(cachedResult);
+					if (parsedResult.tokenUsage) parsedResult.tokenUsage.cached = parsedResult.tokenUsage.total || 0;
+					if (isTransformed && parsedResult.metadata) {
+						parsedResult.metadata.transformed = true;
+						parsedResult.metadata.originalPrompt = prompt;
+					}
+					return {
+						...parsedResult,
+						cached: true
+					};
+				} catch (_) {
+					logger.warn(`Failed to parse cached SageMaker response: ${_}`);
+				}
+			}
+		}
+		if (delayMs && delayMs > 0) {
+			logger.debug(`Applying delay of ${delayMs}ms before calling SageMaker endpoint ${this.getEndpointName()}`);
+			await sleep(delayMs);
+		}
+		const runtime = await this.getSageMakerRuntimeInstance();
+		const payload = this.formatPayload(transformedPrompt);
+		logger.debug(`Calling SageMaker endpoint ${this.getEndpointName()}`);
+		logger.debug(`With payload: ${payload.length > 1e3 ? payload.substring(0, 1e3) + "..." : payload}`);
+		try {
+			const { InvokeEndpointCommand } = await import("@aws-sdk/client-sagemaker-runtime");
+			const command = new InvokeEndpointCommand({
+				EndpointName: this.getEndpointName(),
+				ContentType: this.getContentType(),
+				Accept: this.getAcceptType(),
+				Body: payload
+			});
+			const startTime = Date.now();
+			const response = await runtime.send(command);
+			const _latency = Date.now() - startTime;
+			if (!response.Body) {
+				logger.error("No response body returned from SageMaker endpoint");
+				return { error: "No response body returned from SageMaker endpoint" };
+			}
+			const responseBody = new TextDecoder().decode(response.Body);
+			logger.debug(`SageMaker response (truncated): ${responseBody.length > 1e3 ? responseBody.substring(0, 1e3) + "..." : responseBody}`);
+			const output = await this.parseResponse(responseBody);
+			if (typeof output === "object" && output !== null && "code" in output) {
+				const code = output.code;
+				if (Number.isInteger(code) && code === 424) {
+					const errorMessage = `API Error: 424${output?.message ? ` ${output.message}` : ""}\n${JSON.stringify(output)}`;
+					logger.error(errorMessage);
+					return { error: errorMessage };
+				}
+			}
+			const promptTokens = Math.ceil(payload.length / 4);
+			const completionTokens = Math.ceil((typeof output === "string" ? output.length : 0) / 4);
+			const result = {
+				output,
+				raw: responseBody,
+				tokenUsage: {
+					prompt: promptTokens,
+					completion: completionTokens,
+					total: promptTokens + completionTokens,
+					cached: 0,
+					numRequests: 1
+				},
+				metadata: {
+					latencyMs: _latency,
+					modelType: this.config.modelType || "custom",
+					transformed: isTransformed,
+					originalPrompt: isTransformed ? prompt : void 0
+				}
+			};
+			if (isCacheEnabled() && !bustCache && result.output && !result.error) {
+				const cacheKey = this.getCacheKey(transformedPrompt);
+				const cache = getCache ? getCache() : await import("./cache-BkrqU5Ba.js").then((n) => n.t).then((m) => m.getCache());
+				const resultToCache = JSON.stringify(result);
+				try {
+					await cache.set(cacheKey, resultToCache);
+					logger.debug(`Stored SageMaker response in cache with key: ${cacheKey.substring(0, 100)}...`);
+				} catch (_) {
+					logger.warn(`Failed to store SageMaker response in cache: ${_}`);
+				}
+			}
+			return result;
+		} catch (error) {
+			logger.error(`SageMaker API error: ${error}`);
+			return { error: `SageMaker API error: ${error.message || String(error)}` };
+		}
+	}
+};
+/**
+* Provider for embeddings with SageMaker endpoints
+*/
+var SageMakerEmbeddingProvider = class extends SageMakerGenericProvider {
+	async callApi() {
+		throw new Error("callApi is not implemented for embedding provider. Use callEmbeddingApi instead.");
+	}
+	/**
+	* Generate a consistent cache key for SageMaker embedding requests
+	* Uses crypto.createHash to generate a shorter, more efficient key
+	*/
+	getCacheKey(text) {
+		const configForKey = {
+			endpoint: this.getEndpointName(),
+			modelType: this.config.modelType,
+			contentType: this.getContentType(),
+			acceptType: this.getAcceptType(),
+			region: this.getRegion(),
+			responseFormat: this.config.responseFormat
+		};
+		const configStr = JSON.stringify(configForKey);
+		const textHash = crypto.createHash("sha256").update(text).digest("hex").substring(0, 16);
+		const configHash = crypto.createHash("sha256").update(configStr).digest("hex").substring(0, 8);
+		return `sagemaker:embedding:v1:${this.getEndpointName()}:${textHash}:${configHash}`;
+	}
+	/**
+	* Invoke SageMaker endpoint for embeddings with caching, delay support, and transformations
+	*/
+	async callEmbeddingApi(text, context) {
+		const { isCacheEnabled, getCache } = await import("./cache-BkrqU5Ba.js").then((n) => n.t);
+		const delayMs = context?.originalProvider?.delay || this.delay;
+		const transformedText = await this.applyTransformation(text, context);
+		const isTransformed = transformedText !== text;
+		if (isTransformed) {
+			logger.debug(`Text transformed for SageMaker embedding endpoint ${this.getEndpointName()}`);
+			logger.debug(`Original: ${text.substring(0, 100)}${text.length > 100 ? "..." : ""}`);
+			logger.debug(`Transformed: ${transformedText.substring(0, 100)}${transformedText.length > 100 ? "..." : ""}`);
+		}
+		const bustCache = context?.debug === true;
+		if (isCacheEnabled() && !bustCache) {
+			const cacheKey = this.getCacheKey(transformedText);
+			const cachedResult = await (await getCache ? await getCache() : await import("./cache-BkrqU5Ba.js").then((n) => n.t).then((m) => m.getCache())).get(cacheKey);
+			if (cachedResult) {
+				logger.debug(`Using cached SageMaker embedding response for ${this.getEndpointName()}`);
+				try {
+					const parsedResult = JSON.parse(cachedResult);
+					if (parsedResult.tokenUsage) parsedResult.tokenUsage.cached = parsedResult.tokenUsage.prompt || 0;
+					return {
+						...parsedResult,
+						cached: true
+					};
+				} catch (_) {
+					logger.warn(`Failed to parse cached SageMaker embedding response: ${_}`);
+				}
+			}
+		}
+		if (delayMs && delayMs > 0) {
+			logger.debug(`Applying delay of ${delayMs}ms before calling SageMaker embedding endpoint ${this.getEndpointName()}`);
+			await sleep(delayMs);
+		}
+		const runtime = await this.getSageMakerRuntimeInstance();
+		let payload;
+		const modelType = this.config.modelType || "custom";
+		logger.debug(`Formatting embedding payload for model type: ${modelType}`);
+		switch (modelType) {
+			case "openai":
+				payload = JSON.stringify({
+					input: transformedText,
+					model: "embedding"
+				});
+				break;
+			case "huggingface":
+				payload = JSON.stringify({ inputs: transformedText });
+				break;
+			default:
+				payload = JSON.stringify({
+					input: transformedText,
+					text: transformedText,
+					inputs: transformedText
+				});
+				break;
+		}
+		logger.debug(`Calling SageMaker embedding endpoint ${this.getEndpointName()}`);
+		logger.debug(`With payload: ${payload}`);
+		try {
+			const { InvokeEndpointCommand } = await import("@aws-sdk/client-sagemaker-runtime");
+			const command = new InvokeEndpointCommand({
+				EndpointName: this.getEndpointName(),
+				ContentType: this.getContentType(),
+				Accept: this.getAcceptType(),
+				Body: payload
+			});
+			const startTime = Date.now();
+			const response = await runtime.send(command);
+			Date.now() - startTime;
+			if (!response.Body) {
+				logger.error("No response body returned from SageMaker embedding endpoint");
+				return { error: "No response body returned from SageMaker embedding endpoint" };
+			}
+			const responseBody = new TextDecoder().decode(response.Body);
+			logger.debug(`SageMaker embedding response: ${responseBody.substring(0, 200)}...`);
+			let responseJson;
+			try {
+				responseJson = JSON.parse(responseBody);
+			} catch (_) {
+				return { error: `Failed to parse embedding response as JSON: ${_}` };
+			}
+			const embedding = responseJson.embedding || responseJson.embeddings || responseJson.data?.[0]?.embedding || (Array.isArray(responseJson) ? responseJson[0] : responseJson);
+			if (this.config.responseFormat?.path) try {
+				const pathExpression = this.config.responseFormat.path;
+				const extracted = await this.extractFromPath(responseJson, pathExpression);
+				if (Array.isArray(extracted) && extracted.every((val) => typeof val === "number")) {
+					const result = {
+						embedding: extracted,
+						tokenUsage: {
+							prompt: Math.ceil(text.length / 4),
+							cached: 0,
+							numRequests: 1
+						},
+						metadata: {
+							transformed: isTransformed,
+							originalText: isTransformed ? text : void 0
+						}
+					};
+					await this.cacheEmbeddingResult(result, transformedText, context, isTransformed, isTransformed ? text : void 0);
+					return result;
+				} else logger.warn("Extracted data is not a valid embedding array, trying other extraction methods");
+			} catch (error) {
+				logger.warn(`Failed to extract embedding from path expression: ${this.config.responseFormat.path}, Error: ${error}`);
+				logger.debug(`Response JSON structure: ${JSON.stringify(responseJson).substring(0, 200)}...`);
+			}
+			if (!embedding || !Array.isArray(embedding)) return { error: `Invalid embedding response format. Could not find embedding array in: ${JSON.stringify(responseJson).substring(0, 100)}...` };
+			const result = {
+				embedding,
+				tokenUsage: {
+					prompt: Math.ceil(text.length / 4),
+					cached: 0,
+					numRequests: 1
+				},
+				metadata: {
+					transformed: isTransformed,
+					originalText: isTransformed ? text : void 0
+				}
+			};
+			await this.cacheEmbeddingResult(result, transformedText, context, isTransformed, isTransformed ? text : void 0);
+			return result;
+		} catch (error) {
+			logger.error(`SageMaker embedding API error: ${error}`);
+			return { error: `SageMaker embedding API error: ${error.message || String(error)}` };
+		}
+	}
+	/**
+	* Helper method to cache embedding results
+	*/
+	async cacheEmbeddingResult(result, text, context, isTransformed = false, originalText) {
+		const { isCacheEnabled, getCache } = await import("./cache-BkrqU5Ba.js").then((n) => n.t);
+		const bustCache = context?.debug === true;
+		if (isCacheEnabled() && !bustCache && result.embedding && !result.error) {
+			const cacheKey = this.getCacheKey(text);
+			const cache = await getCache ? await getCache() : await import("./cache-BkrqU5Ba.js").then((n) => n.t).then((m) => m.getCache());
+			if (isTransformed && originalText && !result.metadata) result.metadata = {
+				transformed: true,
+				originalText
+			};
+			else if (isTransformed && originalText && result.metadata) {
+				result.metadata.transformed = true;
+				result.metadata.originalText = originalText;
+			}
+			const resultToCache = JSON.stringify(result);
+			try {
+				await cache.set(cacheKey, resultToCache);
+				logger.debug(`Stored SageMaker embedding response in cache with key: ${cacheKey.substring(0, 100)}...`);
+			} catch (_) {
+				logger.warn(`Failed to store SageMaker embedding response in cache: ${_}`);
+			}
+		}
+	}
+};
+//#endregion
+export { SageMakerCompletionProvider, SageMakerEmbeddingProvider };
+
+//# sourceMappingURL=sagemaker-Cno2V-Sx.js.map