@superlinked/sie-sdk 0.6.20 → 0.6.21

package/dist/index.js

@@ -590,7 +590,11 @@ function parseScoreResult(data) {
   return {
     model: wire.model,
     queryId: wire.query_id,
-    scores: wire.scores.map(parseScoreEntry)
+    scores: wire.scores.map(parseScoreEntry),
+    usage: wire.usage ? {
+      inputTokens: wire.usage.input_tokens,
+      images: wire.usage.images
+    } : void 0
   };
 }
 function parseEntity(data) {
@@ -603,6 +607,19 @@ function parseEntity(data) {
     bbox: data.bbox
   };
 }
+function parseExtractItemError(data) {
+  if (data === void 0 || data === null) return void 0;
+  if (typeof data === "object" && !Array.isArray(data)) {
+    const error = data;
+    if (typeof error.code === "string" && error.code.trim().length > 0 && typeof error.message === "string" && error.message.trim().length > 0) {
+      return { code: error.code, message: error.message };
+    }
+  }
+  return {
+    code: "INTERNAL_ERROR",
+    message: "Malformed extraction item error"
+  };
+}
 function parseExtractResult(data) {
   return {
     id: data.id,
@@ -627,7 +644,9 @@ function parseExtractResult(data) {
         score: o.score,
         bbox: o.bbox
       })
-    )
+    ),
+    data: data.data,
+    error: parseExtractItemError(data.error)
   };
 }
 function parseExtractResults(data) {
@@ -1170,12 +1189,75 @@ function extractDataPayload(block) {
 }
 
 // src/version.ts
-var SDK_VERSION = "0.6.20";
+var SDK_VERSION = "0.6.21";
 
 // src/client.ts
 function sleep2(ms) {
   return new Promise((resolve) => setTimeout(resolve, ms));
 }
+function parseNonnegativeMeterHeader(headers, name) {
+  const raw = headers.get(name);
+  if (raw === null || !/^(0|[1-9]\d*)$/.test(raw)) return void 0;
+  const parsed = Number(raw);
+  return Number.isSafeInteger(parsed) ? parsed : void 0;
+}
+function parseRequestMetadata(headers) {
+  const metadata = {};
+  const requestId = headers.get("x-sie-request-id");
+  if (requestId !== null && requestId.length > 0 && requestId.length <= 256 && requestId === requestId.trim() && /^[\x20-\x7e]+$/.test(requestId)) {
+    metadata.id = requestId;
+  }
+  const executionIdentitySha256 = headers.get("x-sie-execution-identity-sha256");
+  if (executionIdentitySha256 !== null && /^[0-9a-f]{64}$/.test(executionIdentitySha256)) {
+    metadata.executionIdentitySha256 = executionIdentitySha256;
+  }
+  const usageHeaders = {
+    inputTokens: "x-sie-units-input-tokens",
+    pairs: "x-sie-units-pairs",
+    images: "x-sie-units-images",
+    pages: "x-sie-units-pages",
+    outputTokens: "x-sie-units-output-tokens",
+    audioMs: "x-sie-units-audio-ms"
+  };
+  const usage = {};
+  for (const [field, header] of Object.entries(usageHeaders)) {
+    const value = parseNonnegativeMeterHeader(headers, header);
+    if (value !== void 0) usage[field] = value;
+  }
+  if (Object.keys(usage).length > 0) metadata.usage = usage;
+  const creditsDebited = parseNonnegativeMeterHeader(headers, "x-sie-credits-debited");
+  if (creditsDebited !== void 0) metadata.creditsDebited = creditsDebited;
+  return Object.keys(metadata).length > 0 ? metadata : void 0;
+}
+function attachRequestMetadata(results, headers) {
+  const metadata = parseRequestMetadata(headers);
+  if (metadata === void 0) return;
+  for (const result of results) {
+    result.request = { ...metadata };
+    if (metadata.usage !== void 0) result.request.usage = { ...metadata.usage };
+  }
+}
+function validateGenerationSeed(seed) {
+  if (!Number.isSafeInteger(seed)) {
+    throw new RangeError("seed must be a JavaScript safe integer");
+  }
+  return seed;
+}
+function applyGenerateOptions(body, options) {
+  if (options.temperature !== void 0) body.temperature = options.temperature;
+  if (options.topP !== void 0) body.top_p = options.topP;
+  if (options.adapterOptions !== void 0) body.options = options.adapterOptions;
+  if (options.stop !== void 0) body.stop = options.stop;
+  if (options.frequencyPenalty !== void 0) body.frequency_penalty = options.frequencyPenalty;
+  if (options.presencePenalty !== void 0) body.presence_penalty = options.presencePenalty;
+  if (options.grammar !== void 0) body.grammar = options.grammar;
+  if (options.seed !== void 0) body.seed = validateGenerationSeed(options.seed);
+  if (options.logitBias !== void 0) body.logit_bias = options.logitBias;
+  if (options.routingKey !== void 0) body.routing_key = options.routingKey;
+  if (options.promptCacheKey !== void 0) body.prompt_cache_key = options.promptCacheKey;
+  if (options.safetyIdentifier !== void 0) body.safety_identifier = options.safetyIdentifier;
+  if (options.loraAdapter !== void 0) body.lora_adapter = options.loraAdapter;
+}
 function resolveUploadFilename(file, filename) {
   if (filename) return filename;
   const name = file.name;
@@ -1217,6 +1299,32 @@ async function itemImagesForWire(item) {
 async function itemsImagesForWire(items) {
   return Promise.all(items.map(itemImagesForWire));
 }
+async function extractItemForWire(item) {
+  const { images, audio, ...wireItem } = item;
+  const wireImages = images ? await Promise.all(images.map(imageForWire)) : void 0;
+  if (!audio) {
+    return { ...wireItem, ...wireImages === void 0 ? {} : { images: wireImages } };
+  }
+  if (audio instanceof Uint8Array) {
+    return {
+      ...wireItem,
+      ...wireImages === void 0 ? {} : { images: wireImages },
+      audio: { data: audio }
+    };
+  }
+  const { sampleRate, ...wireAudio } = audio;
+  return {
+    ...wireItem,
+    ...wireImages === void 0 ? {} : { images: wireImages },
+    audio: {
+      ...wireAudio,
+      ...sampleRate === void 0 ? {} : { sample_rate: sampleRate }
+    }
+  };
+}
+async function itemsForExtractWire(items) {
+  return Promise.all(items.map(extractItemForWire));
+}
 function extractChatChunkError(chunk) {
   const err = chunk.error;
   if (!err) return null;
@@ -1341,6 +1449,7 @@ var SIEClient = class {
     );
     const data = unpackMessage(new Uint8Array(await response.arrayBuffer()));
     const results = parseEncodeResults(data.items);
+    attachRequestMetadata(results, response.headers);
     if (isSingleItem) {
       const first = results[0];
       if (!first) {
@@ -1510,13 +1619,9 @@ var SIEClient = class {
   async generate(model, prompt, options) {
     const body = {
       prompt,
-      max_new_tokens: options.maxNewTokens,
-      temperature: options.temperature ?? 1,
-      top_p: options.topP ?? 1
+      max_new_tokens: options.maxNewTokens
     };
-    if (options.stop !== void 0) {
-      body.stop = options.stop;
-    }
+    applyGenerateOptions(body, options);
     const { pool, gpu } = this.parseGpuParam(options.gpu);
     const headers = {
       Accept: "application/json",
@@ -1539,7 +1644,9 @@ var SIEClient = class {
     if (data === null || typeof data !== "object") {
       throw new RequestError("Unexpected generate response shape");
     }
-    return parseGenerateResult(data);
+    const result = parseGenerateResult(data);
+    attachRequestMetadata([result], response.headers);
+    return result;
   }
   /**
    * Per-attempt JSON POST used by the non-streaming surfaces
@@ -1617,6 +1724,9 @@ var SIEClient = class {
         400
       );
     }
+    if (req.seed !== void 0) {
+      validateGenerationSeed(req.seed);
+    }
     const body = { ...req, stream: false };
     const url = `${this.baseUrl}/v1/chat/completions`;
     const headers = this.buildChatHeaders("application/json");
@@ -1633,6 +1743,7 @@ var SIEClient = class {
     if (data === null || typeof data !== "object") {
       throw new RequestError("Unexpected chat.completion response shape");
     }
+    attachRequestMetadata([data], response.headers);
     return data;
   }
   /**
@@ -1684,6 +1795,9 @@ var SIEClient = class {
    * ```
    */
   async *streamChatCompletions(req, signal) {
+    if (req.seed !== void 0) {
+      validateGenerationSeed(req.seed);
+    }
     const body = { ...req, stream: true };
     const url = `${this.baseUrl}/v1/chat/completions`;
     yield* this.consumeSseStream(
@@ -1700,10 +1814,10 @@ var SIEClient = class {
    * chunk shape documented in
    * `packages/sie_gateway/src/handlers/sse.rs::build_generate_chunk_event`.
    *
-   * The first delta carries `seq: 0` and `text_delta` populated; the
-   * terminal chunk has `done: true`, `finish_reason`, and (typically)
-   * `usage` + `ttft_ms`. The generator completes on the `data: [DONE]`
-   * sentinel.
+   * Delta chunks may carry text, log probabilities, or both; callers must not
+   * assume every non-terminal event has a non-empty `text_delta`. The terminal
+   * chunk has `done: true`, `finish_reason`, and (typically) `usage` +
+   * `ttft_ms`. The generator completes on the `data: [DONE]` sentinel.
    *
    * Error semantics match {@link streamChatCompletions}: pre-stream HTTP
    * errors throw normally, mid-stream `error` chunks throw
@@ -1725,11 +1839,14 @@ var SIEClient = class {
     const body = {
       prompt,
       max_new_tokens: options.maxNewTokens,
-      temperature: options.temperature ?? 1,
-      top_p: options.topP ?? 1,
       stream: true
     };
-    if (options.stop !== void 0) body.stop = options.stop;
+    applyGenerateOptions(body, options);
+    if (options.logprobs !== void 0) body.logprobs = options.logprobs;
+    if (options.topLogprobs !== void 0) {
+      body.top_logprobs = options.topLogprobs;
+      body.logprobs = true;
+    }
     const safeModel = model.replaceAll("/", "__");
     const url = `${this.baseUrl}/v1/generate/${encodeURIComponent(safeModel)}`;
     const { pool, gpu } = this.parseGpuParam(options.gpu);
@@ -1946,7 +2063,9 @@ var SIEClient = class {
       model
     );
     const data = unpackMessage(new Uint8Array(await response.arrayBuffer()));
-    return parseScoreResult(data);
+    const result = parseScoreResult(data);
+    attachRequestMetadata([result], response.headers);
+    return result;
   }
   /**
    * Extract entities from one or more items.
@@ -1970,7 +2089,7 @@ var SIEClient = class {
   async extract(model, items, options) {
     const isSingleItem = !Array.isArray(items);
     const itemsArray = isSingleItem ? [items] : items;
-    const itemsForWire = await itemsImagesForWire(itemsArray);
+    const itemsForWire = await itemsForExtractWire(itemsArray);
     const body = {
       items: itemsForWire
     };
@@ -1996,6 +2115,7 @@ var SIEClient = class {
     );
     const data = unpackMessage(new Uint8Array(await response.arrayBuffer()));
     const results = parseExtractResults(data.items);
+    attachRequestMetadata(results, response.headers);
     if (isSingleItem) {
       const first = results[0];
       if (!first) {