@bitfab/sdk 0.18.0 → 0.18.1

package/dist/index.cjs

@@ -433,7 +433,7 @@ __export(index_exports, {
 module.exports = __toCommonJS(index_exports);
 
 // src/version.generated.ts
-var __version__ = "0.18.0";
+var __version__ = "0.18.1";
 
 // src/constants.ts
 var DEFAULT_SERVICE_URL = "https://bitfab.ai";
@@ -1663,21 +1663,131 @@ function extractModelName(serialized, metadata) {
   }
   return void 0;
 }
+function asTokenCount(value) {
+  return typeof value === "number" && Number.isFinite(value) ? value : null;
+}
+function normalizeTokenUsage(raw) {
+  if (typeof raw !== "object" || raw === null || Array.isArray(raw)) {
+    return null;
+  }
+  const u = raw;
+  if ("cache_read_input_tokens" in u || "cache_creation_input_tokens" in u) {
+    const cacheRead = asTokenCount(u.cache_read_input_tokens);
+    const cacheCreation = asTokenCount(u.cache_creation_input_tokens);
+    const baseInput = asTokenCount(u.input_tokens);
+    const outputTokens = asTokenCount(u.output_tokens);
+    if (cacheRead === null && cacheCreation === null && baseInput === null && outputTokens === null) {
+      return null;
+    }
+    const inputTokens = (baseInput ?? 0) + (cacheRead ?? 0) + (cacheCreation ?? 0);
+    return {
+      inputTokens,
+      outputTokens,
+      totalTokens: inputTokens + (outputTokens ?? 0),
+      cachedInputTokens: cacheRead
+    };
+  }
+  if ("prompt_tokens" in u || "completion_tokens" in u || "promptTokens" in u || "completionTokens" in u) {
+    const promptDetails = u.prompt_tokens_details ?? {};
+    return withAnyTokenCount({
+      inputTokens: asTokenCount(u.prompt_tokens) ?? asTokenCount(u.promptTokens),
+      outputTokens: asTokenCount(u.completion_tokens) ?? asTokenCount(u.completionTokens),
+      totalTokens: asTokenCount(u.total_tokens) ?? asTokenCount(u.totalTokens),
+      cachedInputTokens: asTokenCount(promptDetails.cached_tokens)
+    });
+  }
+  if ("prompt_token_count" in u || "candidates_token_count" in u) {
+    return withAnyTokenCount({
+      inputTokens: asTokenCount(u.prompt_token_count),
+      outputTokens: asTokenCount(u.candidates_token_count),
+      totalTokens: asTokenCount(u.total_token_count),
+      cachedInputTokens: asTokenCount(u.cached_content_token_count)
+    });
+  }
+  if ("input_tokens" in u || "output_tokens" in u) {
+    const inputDetails = u.input_token_details ?? {};
+    const inputTokens = asTokenCount(u.input_tokens);
+    const outputTokens = asTokenCount(u.output_tokens);
+    let totalTokens = asTokenCount(u.total_tokens);
+    if (totalTokens === null && inputTokens !== null && outputTokens !== null) {
+      totalTokens = inputTokens + outputTokens;
+    }
+    return withAnyTokenCount({
+      inputTokens,
+      outputTokens,
+      totalTokens,
+      cachedInputTokens: asTokenCount(inputDetails.cache_read)
+    });
+  }
+  return null;
+}
+function withAnyTokenCount(usage) {
+  const hasCount = usage.inputTokens !== null || usage.outputTokens !== null || usage.totalTokens !== null || usage.cachedInputTokens !== null;
+  return hasCount ? usage : null;
+}
+function addUsage(totals, usage) {
+  for (const key of [
+    "inputTokens",
+    "outputTokens",
+    "totalTokens",
+    "cachedInputTokens"
+  ]) {
+    const value = usage[key];
+    if (value !== null) {
+      totals[key] = (totals[key] ?? 0) + value;
+    }
+  }
+}
+function usageFromGenerations(generations) {
+  if (!generations?.length) {
+    return null;
+  }
+  const totals = {
+    inputTokens: null,
+    outputTokens: null,
+    totalTokens: null,
+    cachedInputTokens: null
+  };
+  let found = false;
+  for (const batch of generations) {
+    if (!Array.isArray(batch)) {
+      continue;
+    }
+    for (const gen of batch) {
+      const msg = gen?.message;
+      if (!msg || typeof msg !== "object") {
+        continue;
+      }
+      const responseMetadata = msg.response_metadata;
+      const usage = normalizeTokenUsage(msg.usage_metadata) ?? normalizeTokenUsage(responseMetadata?.token_usage) ?? normalizeTokenUsage(responseMetadata?.usage) ?? normalizeTokenUsage(responseMetadata?.tokenUsage);
+      if (!usage) {
+        continue;
+      }
+      found = true;
+      addUsage(totals, usage);
+    }
+  }
+  return found ? totals : null;
+}
 function extractUsage2(output) {
+  const generations = output.generations;
+  const llmOutput = output.llmOutput ?? output.llm_output;
+  const normalized = usageFromGenerations(generations) ?? normalizeTokenUsage(llmOutput?.tokenUsage) ?? normalizeTokenUsage(llmOutput?.token_usage) ?? normalizeTokenUsage(llmOutput?.usage);
   const usage = {};
-  const llmOutput = output.llmOutput;
-  const tokenUsage = llmOutput?.tokenUsage ?? llmOutput?.token_usage ?? llmOutput?.usage ?? {};
-  const inputTokens = tokenUsage.promptTokens ?? tokenUsage.prompt_tokens ?? tokenUsage.input_tokens;
-  const outputTokens = tokenUsage.completionTokens ?? tokenUsage.completion_tokens ?? tokenUsage.output_tokens;
-  const totalTokens = tokenUsage.totalTokens ?? tokenUsage.total_tokens;
-  if (inputTokens !== void 0 && inputTokens !== null) {
-    usage.inputTokens = inputTokens;
-  }
-  if (outputTokens !== void 0 && outputTokens !== null) {
-    usage.outputTokens = outputTokens;
-  }
-  if (totalTokens !== void 0 && totalTokens !== null) {
-    usage.totalTokens = totalTokens;
+  if (!normalized) {
+    return usage;
+  }
+  if (normalized.inputTokens !== null) {
+    usage.inputTokens = normalized.inputTokens;
+  }
+  if (normalized.outputTokens !== null) {
+    usage.outputTokens = normalized.outputTokens;
+  }
+  if (normalized.totalTokens !== null) {
+    usage.totalTokens = normalized.totalTokens;
+  }
+  if (normalized.cachedInputTokens !== null) {
+    usage.cachedInputTokens = normalized.cachedInputTokens;
   }
   return usage;
 }