@ai-sdk-tool/eval 0.1.3 → 0.1.5

package/dist/index.cjs

@@ -113,7 +113,7 @@ function uniqueLines(lines) {
 function suggestFixFromDiff(parsed) {
   const suggestions = [];
   const { error_type, expected, actual, diff } = parsed ?? {};
-  if (diff && diff.some((d) => d.includes("function name")) || diff && diff.some((d) => d.includes("missing function:"))) {
+  if (Array.isArray(diff) && diff.some((d) => String(d).includes("function name")) || Array.isArray(diff) && diff.some((d) => String(d).includes("missing function:"))) {
     const expectedName = expected?.function;
     const actualName = actual?.function;
     if (expectedName && actualName && expectedName !== actualName) {
@@ -127,23 +127,23 @@ function suggestFixFromDiff(parsed) {
       );
     }
   }
-  if (diff && diff.some((d) => d.startsWith("- missing required param:"))) {
-    const missing = diff.filter((d) => d.startsWith("- missing required param:")).map((d) => d.replace("- missing required param: ", ""));
+  if (Array.isArray(diff) && diff.some((d) => String(d).startsWith("- missing required param:"))) {
+    const missing = diff.filter((d) => String(d).startsWith("- missing required param:")).map((d) => String(d).replace("- missing required param: ", ""));
     if (missing.length) {
       suggestions.push(`Add required parameter(s): ${missing.join(", ")}.`);
     }
   }
-  if (diff && diff.some((d) => d.startsWith("+ unexpected param:"))) {
-    const extras = diff.filter((d) => d.startsWith("+ unexpected param:")).map((d) => d.replace("+ unexpected param: ", ""));
+  if (Array.isArray(diff) && diff.some((d) => String(d).startsWith("+ unexpected param:"))) {
+    const extras = diff.filter((d) => String(d).startsWith("+ unexpected param:")).map((d) => String(d).replace("+ unexpected param: ", ""));
     if (extras.length) {
       suggestions.push(`Remove unexpected parameter(s): ${extras.join(", ")}.`);
     }
   }
-  if (diff && diff.some((d) => d.startsWith("@@ param "))) {
-    const targets = diff.filter((d) => d.startsWith("@@ param ")).map((d) => d.replace("@@ param ", ""));
+  if (Array.isArray(diff) && diff.some((d) => String(d).startsWith("@@ param "))) {
+    const targets = diff.filter((d) => String(d).startsWith("@@ param ")).map((d) => String(d).replace("@@ param ", ""));
     for (const param of targets) {
       const allowedLine = diff.find(
-        (d) => d.startsWith("- expected one of:")
+        (d) => String(d).startsWith("- expected one of:")
       );
       if (allowedLine) {
         const allowed = allowedLine.replace("- expected one of: ", "");
@@ -281,13 +281,13 @@ var reporters = {
 };
 
 // src/evaluate.ts
-async function runSingleBenchmark(model, benchmark, modelKey) {
+async function runSingleBenchmark(model, benchmark, modelKey, config) {
   const modelId = typeof model === "object" && model !== null && "modelId" in model && typeof model.modelId === "string" ? model.modelId : "unknown-model";
   try {
     console.log(
       `[${modelId}]${modelKey ? ` (${modelKey})` : ""} Running benchmark: ${benchmark.name}...`
     );
-    const result = await benchmark.run(model);
+    const result = await benchmark.run(model, config);
     console.log(
       `[${modelId}]${modelKey ? ` (${modelKey})` : ""} Finished benchmark: ${benchmark.name}. Score: ${result.score}`
     );
@@ -316,7 +316,7 @@ async function runSingleBenchmark(model, benchmark, modelKey) {
   }
 }
 async function evaluate(options) {
-  const { models, benchmarks, reporter = "console" } = options;
+  const { models, benchmarks, reporter = "console", temperature } = options;
   const modelEntries = [];
   if (Array.isArray(models)) {
     for (const m of models) modelEntries.push([void 0, m]);
@@ -335,7 +335,8 @@ async function evaluate(options) {
       const evaluationResult = await runSingleBenchmark(
         model,
         benchmark,
-        modelKey
+        modelKey,
+        temperature !== void 0 ? { temperature } : void 0
       );
       allResults.push(evaluationResult);
     }
@@ -350,17 +351,16 @@ async function evaluate(options) {
   return allResults;
 }
 
-// src/benchmarks/json-generation.ts
+// src/benchmarks/bfcl.ts
 var import_ai = require("ai");
-var import_ajv = __toESM(require("ajv"), 1);
 var import_fs2 = require("fs");
 var import_path2 = __toESM(require("path"), 1);
 
 // src/utils/paths.ts
 var import_fs = __toESM(require("fs"), 1);
+var import_module = require("module");
 var import_path = __toESM(require("path"), 1);
 var import_url = require("url");
-var import_module = require("module");
 function resolveDataDir(fromModuleUrl) {
   const moduleUrl = fromModuleUrl;
   const override = process.env.BFCL_DATA_DIR;
@@ -408,263 +408,6 @@ function resolveDataDir(fromModuleUrl) {
   return import_path.default.join(pkgRoot, "data");
 }
 
-// src/benchmarks/json-generation.ts
-function extractFirstJsonBlock(text) {
-  try {
-    return JSON.parse(text);
-  } catch {
-  }
-  const fenceMatch = text.match(/```json\s*([\s\S]*?)```/i) || text.match(/```\s*([\s\S]*?)```/i);
-  if (fenceMatch) {
-    const inner = fenceMatch[1].trim();
-    try {
-      return JSON.parse(inner);
-    } catch {
-    }
-  }
-  const startIdxObj = text.indexOf("{");
-  const startIdxArr = text.indexOf("[");
-  const start = [startIdxObj, startIdxArr].filter((i) => i >= 0).sort((a, b) => a - b)[0];
-  if (start === void 0) return void 0;
-  const open = text[start] === "{" ? "{" : "[";
-  const close = open === "{" ? "}" : "]";
-  let depth = 0;
-  for (let i = start; i < text.length; i++) {
-    const ch = text[i];
-    if (ch === open) depth++;
-    else if (ch === close) depth--;
-    if (depth === 0) {
-      const candidate = text.slice(start, i + 1);
-      try {
-        return JSON.parse(candidate);
-      } catch {
-      }
-      break;
-    }
-  }
-  return void 0;
-}
-function subsetMatch(expected, actual) {
-  if (expected === null || typeof expected !== "object") {
-    return expected === actual;
-  }
-  if (Array.isArray(expected)) {
-    if (!Array.isArray(actual)) return false;
-    for (let i = 0; i < expected.length; i++) {
-      if (!subsetMatch(expected[i], actual[i])) return false;
-    }
-    return true;
-  }
-  if (actual === null || typeof actual !== "object") return false;
-  const eObj = expected;
-  const aObj = actual;
-  for (const key of Object.keys(eObj)) {
-    if (!subsetMatch(eObj[key], aObj[key])) return false;
-  }
-  return true;
-}
-var jsonGenerationBenchmark = {
-  name: "json-generation",
-  version: "2.1.0",
-  description: "Evaluates schema-compliant JSON generation from natural language using JSON Schema prompts.",
-  async run(model) {
-    const logs = [];
-    const ajv = new import_ajv.default({ allErrors: true, strict: false });
-    let schemaValidCount = 0;
-    let valueMatchCount = 0;
-    let correctCount = 0;
-    let tests = [];
-    const expectedMap = /* @__PURE__ */ new Map();
-    try {
-      const dataDir = resolveDataDir();
-      const testsJsonl = await import_fs2.promises.readFile(
-        import_path2.default.join(dataDir, "json_generation_tests.jsonl"),
-        "utf-8"
-      );
-      const expectedJsonl = await import_fs2.promises.readFile(
-        import_path2.default.join(dataDir, "json_generation_expected.jsonl"),
-        "utf-8"
-      );
-      tests = testsJsonl.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
-      const expecteds = expectedJsonl.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
-      for (const r of expecteds) expectedMap.set(r.id, r);
-    } catch (e) {
-      const msg = e instanceof Error ? e.message : String(e);
-      return {
-        score: 0,
-        success: false,
-        metrics: {},
-        logs: [`[FATAL] Failed to load json-generation datasets: ${msg}`],
-        error: e
-      };
-    }
-    for (const tc of tests) {
-      try {
-        const schemaStr = JSON.stringify(tc.schema, null, 2);
-        const messages = [
-          {
-            role: "system",
-            content: "You must output only a single JSON document that strictly conforms to the given JSON Schema. Do not include any extra text or code fences."
-          },
-          {
-            role: "user",
-            content: [
-              "Generate a JSON object that reflects the following facts.",
-              "JSON Schema:",
-              schemaStr,
-              "Facts:",
-              tc.promptFacts,
-              "Output must be a single JSON only, with no additional text."
-            ].join("\n\n")
-          }
-        ];
-        const { text } = await (0, import_ai.generateText)({ model, messages });
-        let parsed;
-        try {
-          parsed = extractFirstJsonBlock(text);
-        } catch {
-        }
-        if (parsed === void 0) {
-          logs.push(`[FAIL] ${tc.id}: Unable to parse JSON from model output.`);
-          continue;
-        }
-        const validate = ajv.compile(tc.schema);
-        const valid = validate(parsed);
-        if (valid) schemaValidCount++;
-        else
-          logs.push(
-            `[INFO] ${tc.id}: Schema validation errors: ${(validate.errors || []).map((e) => `${e.instancePath} ${e.message}`).join(", ") || "unknown"}`
-          );
-        const expectedRec = expectedMap.get(tc.id);
-        if (!expectedRec) {
-          logs.push(
-            `[WARN] ${tc.id}: No expected record found. Skipping value match.`
-          );
-        }
-        const valuesOk = expectedRec ? subsetMatch(expectedRec.expected, parsed) : false;
-        if (valuesOk) valueMatchCount++;
-        if (valid && valuesOk) {
-          correctCount++;
-          logs.push(`[PASS] ${tc.id}`);
-        } else {
-          logs.push(
-            `[FAIL] ${tc.id}: schemaValid=${valid}, valuesOk=${valuesOk}. Output=${JSON.stringify(
-              parsed
-            )}`
-          );
-        }
-      } catch (e) {
-        const msg = e instanceof Error ? e.message : String(e);
-        logs.push(`[ERROR] ${tc.id}: ${msg}`);
-      }
-    }
-    const total = tests.length;
-    const score = correctCount / total;
-    return {
-      score,
-      success: score >= 0.8,
-      metrics: {
-        total_cases: total,
-        correct_count: correctCount,
-        schema_valid_count: schemaValidCount,
-        value_match_count: valueMatchCount,
-        accuracy: score
-      },
-      logs
-    };
-  }
-};
-var jsonGenerationSchemaOnlyBenchmark = {
-  name: "json-generation-schema-only",
-  version: "1.0.1",
-  description: "Evaluates whether model outputs strictly conform to the provided JSON Schema (structure only).",
-  async run(model) {
-    const logs = [];
-    const ajv = new import_ajv.default({ allErrors: true, strict: false });
-    let tests = [];
-    try {
-      const dataDir = resolveDataDir();
-      const testsJsonl = await import_fs2.promises.readFile(
-        import_path2.default.join(dataDir, "json_generation_tests.jsonl"),
-        "utf-8"
-      );
-      tests = testsJsonl.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
-    } catch (e) {
-      const msg = e instanceof Error ? e.message : String(e);
-      return {
-        score: 0,
-        success: false,
-        metrics: {},
-        logs: [`[FATAL] Failed to load schema-only tests: ${msg}`],
-        error: e
-      };
-    }
-    let schemaValidCount = 0;
-    for (const tc of tests) {
-      try {
-        const schemaStr = JSON.stringify(tc.schema, null, 2);
-        const messages = [
-          {
-            role: "system",
-            content: "You must output only a single JSON document that strictly conforms to the given JSON Schema. Do not include any extra text or code fences."
-          },
-          {
-            role: "user",
-            content: [
-              "Generate a JSON object that reflects the following facts.",
-              "JSON Schema:",
-              schemaStr,
-              "Facts:",
-              tc.promptFacts,
-              "Output must be a single JSON only, with no additional text."
-            ].join("\n\n")
-          }
-        ];
-        const { text } = await (0, import_ai.generateText)({ model, messages });
-        let parsed;
-        try {
-          parsed = extractFirstJsonBlock(text);
-        } catch {
-        }
-        if (parsed === void 0) {
-          logs.push(`[FAIL] ${tc.id}: Could not parse JSON from model output.`);
-          continue;
-        }
-        const validate = ajv.compile(tc.schema);
-        const valid = validate(parsed);
-        if (valid) {
-          schemaValidCount++;
-          logs.push(`[PASS] ${tc.id}`);
-        } else {
-          logs.push(
-            `[FAIL] ${tc.id}: Schema validation errors: ${(validate.errors || []).map((e) => `${e.instancePath} ${e.message}`).join(", ") || "unknown"}`
-          );
-        }
-      } catch (e) {
-        const msg = e instanceof Error ? e.message : String(e);
-        logs.push(`[ERROR] ${tc.id}: ${msg}`);
-      }
-    }
-    const total = tests.length;
-    const score = total > 0 ? schemaValidCount / total : 0;
-    return {
-      score,
-      success: score >= 0.8,
-      metrics: {
-        total_cases: total,
-        schema_valid_count: schemaValidCount,
-        accuracy: score
-      },
-      logs
-    };
-  }
-};
-
-// src/benchmarks/bfcl.ts
-var import_ai2 = require("ai");
-var import_fs3 = require("fs");
-var import_path3 = __toESM(require("path"), 1);
-
 // src/benchmarks/bfcl/ast-checker.ts
 function standardizeString(input) {
   if (typeof input !== "string") return input;
@@ -674,7 +417,7 @@ function standardizeString(input) {
 function checkStringValue(param, modelValue, possibleAnswers) {
   const standardizedModelValue = standardizeString(modelValue);
   const standardizedPossibleAnswers = possibleAnswers.map(
-    (ans) => standardizeString(ans)
+    (ans) => standardizeString(String(ans))
   );
   if (!standardizedPossibleAnswers.includes(standardizedModelValue)) {
     return {
@@ -701,8 +444,9 @@ function simpleFunctionChecker(funcDescription, modelToolCall, possibleAnswer) {
     };
   }
   const possibleAnswerParams = possibleAnswer[Object.keys(possibleAnswer)[0]];
+  const argsObj = modelArgs && typeof modelArgs === "object" ? modelArgs : {};
   for (const param of requiredParams) {
-    if (!(param in modelArgs)) {
+    if (!(param in argsObj)) {
       return {
         valid: false,
         error: `Missing required parameter: '${param}'.`,
@@ -710,87 +454,98 @@ function simpleFunctionChecker(funcDescription, modelToolCall, possibleAnswer) {
       };
     }
   }
-  for (const paramName in modelArgs) {
-    const modelValue = modelArgs[paramName];
-    if (!(paramName in expectedParams) || !(paramName in possibleAnswerParams)) {
-      return {
-        valid: false,
-        error: `Unexpected parameter: '${paramName}'.`,
-        error_type: "simple_function_checker:unexpected_param"
-      };
-    }
-    const possibleValues = possibleAnswerParams[paramName];
-    if (typeof modelValue === "string") {
-      const result = checkStringValue(paramName, modelValue, possibleValues);
-      if (!result.valid) return result;
-    } else if (Array.isArray(modelValue)) {
-      const modelValueStr = JSON.stringify(
-        modelValue.map((v) => standardizeString(v.toString())).sort()
-      );
-      const hasMatch = possibleValues.some(
-        (p) => JSON.stringify(
-          p.map((v) => standardizeString(v.toString())).sort()
-        ) === modelValueStr
-      );
-      if (!hasMatch) {
+  if (modelArgs && typeof modelArgs === "object") {
+    for (const paramName of Object.keys(argsObj)) {
+      const modelValue = argsObj[paramName];
+      if (!(paramName in expectedParams) || !(paramName in possibleAnswerParams)) {
         return {
           valid: false,
-          error: `Invalid value for list parameter '${paramName}'. Got ${JSON.stringify(
-            modelValue
-          )}. Expected one of ${JSON.stringify(possibleValues)}.`,
-          error_type: "value_error:list"
+          error: `Unexpected parameter: '${paramName}'.`,
+          error_type: "simple_function_checker:unexpected_param"
         };
       }
-    } else {
-      const hasMatch = possibleValues.some((possibleValue) => {
-        if (modelValue === possibleValue) return true;
-        if (typeof modelValue === "object" && modelValue !== null && typeof possibleValue === "object" && possibleValue !== null) {
-          try {
-            const normalizeObject = (obj) => {
-              if (Array.isArray(obj)) {
-                return obj.map(normalizeObject);
-              }
-              if (obj && typeof obj === "object") {
-                const normalized = {};
-                for (const [key, value] of Object.entries(obj)) {
-                  if (Array.isArray(value) && value.length === 1 && (typeof value[0] !== "object" || value[0] === null)) {
-                    normalized[key] = value[0];
-                  } else {
-                    normalized[key] = normalizeObject(value);
+      const possibleValues = possibleAnswerParams[paramName];
+      if (typeof modelValue === "string") {
+        const result = checkStringValue(
+          paramName,
+          modelValue,
+          possibleValues ?? []
+        );
+        if (!result.valid) return result;
+      } else if (Array.isArray(modelValue)) {
+        const modelValueStr = JSON.stringify(
+          modelValue.map((v) => standardizeString(String(v))).sort()
+        );
+        const hasMatch = Array.isArray(possibleValues) ? possibleValues.some((p) => {
+          if (!Array.isArray(p)) return false;
+          return JSON.stringify(
+            p.map((v) => standardizeString(String(v))).sort()
+          ) === modelValueStr;
+        }) : false;
+        if (!hasMatch) {
+          return {
+            valid: false,
+            error: `Invalid value for list parameter '${paramName}'. Got ${JSON.stringify(
+              modelValue
+            )}. Expected one of ${JSON.stringify(possibleValues)}.`,
+            error_type: "value_error:list"
+          };
+        }
+      } else {
+        const hasMatch = Array.isArray(possibleValues) ? possibleValues.some((possibleValue) => {
+          if (modelValue === possibleValue) return true;
+          if (typeof modelValue === "object" && modelValue !== null && typeof possibleValue === "object" && possibleValue !== null) {
+            try {
+              const normalizeObject = (obj) => {
+                if (Array.isArray(obj)) {
+                  return obj.map(normalizeObject);
+                }
+                if (obj && typeof obj === "object") {
+                  const normalized = {};
+                  for (const [key, value] of Object.entries(
+                    obj
+                  )) {
+                    if (Array.isArray(value) && value.length === 1 && (typeof value[0] !== "object" || value[0] === null)) {
+                      normalized[key] = value[0];
+                    } else {
+                      normalized[key] = normalizeObject(value);
+                    }
                   }
+                  return normalized;
                 }
-                return normalized;
-              }
-              return obj;
-            };
-            const normalizedModel = normalizeObject(modelValue);
-            const normalizedPossible = normalizeObject(possibleValue);
-            return JSON.stringify(normalizedModel) === JSON.stringify(normalizedPossible);
-          } catch {
-            return false;
+                return obj;
+              };
+              const normalizedModel = normalizeObject(modelValue);
+              const normalizedPossible = normalizeObject(possibleValue);
+              return JSON.stringify(normalizedModel) === JSON.stringify(normalizedPossible);
+            } catch {
+              return false;
+            }
           }
+          if (typeof modelValue === "number" && typeof possibleValue === "string") {
+            return modelValue.toString() === possibleValue;
+          }
+          if (typeof modelValue === "string" && typeof possibleValue === "number") {
+            return modelValue === possibleValue.toString();
+          }
+          return false;
+        }) : false;
+        if (!hasMatch) {
+          return {
+            valid: false,
+            error: `Invalid value for parameter '${paramName}'. Got ${JSON.stringify(
+              modelValue
+            )}. Expected one of ${JSON.stringify(possibleValues)}.`,
+            error_type: "value_error:other"
+          };
         }
-        if (typeof modelValue === "number" && typeof possibleValue === "string") {
-          return modelValue.toString() === possibleValue;
-        }
-        if (typeof modelValue === "string" && typeof possibleValue === "number") {
-          return modelValue === possibleValue.toString();
-        }
-        return false;
-      });
-      if (!hasMatch) {
-        return {
-          valid: false,
-          error: `Invalid value for parameter '${paramName}'. Got ${JSON.stringify(
-            modelValue
-          )}. Expected one of ${JSON.stringify(possibleValues)}.`,
-          error_type: "value_error:other"
-        };
       }
     }
   }
   for (const paramName in possibleAnswerParams) {
-    if (!(paramName in modelArgs) && !possibleAnswerParams[paramName].includes("")) {
+    const val = possibleAnswerParams[paramName];
+    const isOptional = Array.isArray(val) && val.includes("");
+    if (!(paramName in argsObj) && !isOptional) {
       return {
         valid: false,
         error: `Missing optional parameter '${paramName}' which was not marked as optional.`,
@@ -876,10 +631,10 @@ function check(testCase, modelOutput, possibleAnswer) {
   const category = testCase.id.split("_")[0];
   try {
     if (category === "simple") {
-      if (!modelOutput || modelOutput.length !== 1) {
+      if (!Array.isArray(modelOutput) || modelOutput.length !== 1) {
         return {
           valid: false,
-          error: `Expected 1 function call, but got ${modelOutput?.length ?? 0}.`,
+          error: `Expected 1 function call, but got ${Array.isArray(modelOutput) ? modelOutput.length : 0}.`,
           error_type: "simple:wrong_count"
         };
       }
@@ -921,19 +676,19 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
     name,
     version: "1.0.0",
     description,
-    async run(model) {
+    async run(model, config) {
       const logs = [];
       let correctCount = 0;
       let testCases = [];
       try {
         const dataPath = resolveDataDir();
         logs.push(`[INFO] Using data dir: ${dataPath}`);
-        const testCasesJson = await import_fs3.promises.readFile(
-          import_path3.default.join(dataPath, testDataFile),
+        const testCasesJson = await import_fs2.promises.readFile(
+          import_path2.default.join(dataPath, testDataFile),
           "utf-8"
         );
-        const possibleAnswersJson = await import_fs3.promises.readFile(
-          import_path3.default.join(dataPath, answerDataFile),
+        const possibleAnswersJson = await import_fs2.promises.readFile(
+          import_path2.default.join(dataPath, answerDataFile),
           "utf-8"
         );
         testCases = testCasesJson.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
@@ -950,19 +705,25 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
           );
         }
         const fixSchema = (schema) => {
-          if (!schema || typeof schema !== "object") return schema;
+          if (!schema || typeof schema !== "object")
+            return { type: "object", properties: {} };
           const copy = Array.isArray(schema) ? schema.map((v) => fixSchema(v)) : { ...schema };
-          if (copy.type) {
-            if (copy.type === "dict") copy.type = "object";
-            if (copy.type === "integer" || copy.type === "float")
-              copy.type = "number";
-          }
-          if (copy.properties && typeof copy.properties === "object") {
-            for (const k of Object.keys(copy.properties)) {
-              copy.properties[k] = fixSchema(copy.properties[k]);
+          if (!Array.isArray(copy)) {
+            if (copy.type) {
+              if (copy.type === "dict") copy.type = "object";
+              if (copy.type === "integer" || copy.type === "float")
+                copy.type = "number";
+            }
+            if (copy.properties && typeof copy.properties === "object") {
+              for (const k of Object.keys(copy.properties)) {
+                copy.properties[k] = fixSchema(
+                  copy.properties[k]
+                );
+              }
             }
+            if (copy.items) copy.items = fixSchema(copy.items);
+            return copy;
           }
-          if (copy.items) copy.items = fixSchema(copy.items);
           return copy;
         };
         const concurrencyEnv = process.env.BFCL_CONCURRENCY;
@@ -973,6 +734,8 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
         const runSingleCase = async (testCase) => {
           const caseLogs = [];
           const { function: tools, question: messages } = testCase;
+          const temp = config?.temperature;
+          const temperature = typeof temp === "number" ? temp : void 0;
           try {
             const flatMessages = Array.isArray(messages) && messages.some((m) => Array.isArray(m)) ? messages.flat(1) : messages;
             const nameMap = /* @__PURE__ */ new Map();
@@ -982,7 +745,8 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
             };
             const transformedTools = tools.map((t) => {
               const fixed = fixSchema(t.parameters);
-              const inputSchema = fixed && typeof fixed === "object" && fixed.type === "object" ? fixed : { type: "object", properties: {} };
+              const isObjectSchema = fixed && typeof fixed === "object" && fixed.type === "object";
+              const inputSchema = isObjectSchema ? fixed : { type: "object", properties: {} };
               const sanitized = sanitizeName(t.name);
               nameMap.set(sanitized, t.name);
               return {
@@ -995,9 +759,9 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
             const toolsMap = Object.fromEntries(
               transformedTools.map((t) => [
                 t.name,
-                (0, import_ai2.tool)({
+                (0, import_ai.tool)({
                   description: typeof t.description === "string" ? t.description : void 0,
-                  inputSchema: (0, import_ai2.jsonSchema)(t.inputSchema)
+                  inputSchema: (0, import_ai.jsonSchema)(t.inputSchema)
                 })
               ])
             );
@@ -1012,16 +776,20 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
                 `[DEBUG] ${testCase.id}: failed to introspect tools: ${e.message}`
               );
             }
-            const { toolCalls, text, finishReason } = await (0, import_ai2.generateText)({
+            const { toolCalls, text, finishReason } = await (0, import_ai.generateText)({
               model,
               messages: flatMessages,
               tools: toolsMap,
               toolChoice: "auto",
+              ...temperature !== void 0 ? { temperature } : {},
               // Pass original schema information to middleware
               providerOptions: {
                 toolCallMiddleware: {
                   originalToolSchemas: Object.fromEntries(
-                    transformedTools.map((t) => [t.name, t.inputSchema])
+                    transformedTools.map((t) => [
+                      t.name,
+                      t.inputSchema
+                    ])
                   )
                 }
               }
@@ -1074,10 +842,13 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
                 const summarizeArgs = (args) => {
                   if (args == null) return args;
                   if (typeof args !== "object") return args;
-                  return Object.keys(args).sort().reduce((acc, k) => {
-                    acc[k] = args[k];
-                    return acc;
-                  }, {});
+                  return Object.keys(args).sort().reduce(
+                    (acc, k) => {
+                      acc[k] = args[k];
+                      return acc;
+                    },
+                    {}
+                  );
                 };
                 const expected = {};
                 const actual = {};
@@ -1098,19 +869,23 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
                     diff.push(`- ${expectedFuncName}`);
                     diff.push(`+ ${receivedName}`);
                   }
-                  if (expectedParams && receivedArgs) {
+                  if (expectedParams && receivedArgs && typeof receivedArgs === "object" && receivedArgs !== null) {
                     const required = funcDesc?.parameters?.required ?? [];
                     for (const req of required) {
                       if (!(req in receivedArgs)) {
                         diff.push(`- missing required param: ${req}`);
                       }
                     }
-                    for (const k of Object.keys(receivedArgs)) {
+                    for (const k of Object.keys(
+                      receivedArgs
+                    )) {
                       if (!Object.prototype.hasOwnProperty.call(expectedParams, k)) {
                         diff.push(`+ unexpected param: ${k}`);
                       }
                     }
-                    for (const k of Object.keys(receivedArgs)) {
+                    for (const k of Object.keys(
+                      receivedArgs
+                    )) {
                       if (Object.prototype.hasOwnProperty.call(expectedParams, k)) {
                         const allowed = expectedParams[k];
                         const got = receivedArgs[k];
@@ -1183,13 +958,15 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
                     );
                     const requiredParams = funcDesc?.parameters?.required ?? [];
                     diff.push(`@@ function ${fname}`);
-                    if (expectedParamsAllowed && receivedArgs) {
+                    if (expectedParamsAllowed && receivedArgs && typeof receivedArgs === "object" && receivedArgs !== null) {
                       for (const req of requiredParams) {
                         if (!(req in receivedArgs)) {
                           diff.push(`- missing required param: ${req}`);
                         }
                       }
-                      for (const k of Object.keys(receivedArgs)) {
+                      for (const k of Object.keys(
+                        receivedArgs
+                      )) {
                         if (!Object.prototype.hasOwnProperty.call(
                           expectedParamsAllowed,
                           k
@@ -1197,7 +974,9 @@ function createBfclBenchmark(name, description, testDataFile, answerDataFile) {
                           diff.push(`+ unexpected param: ${k}`);
                         }
                       }
-                      for (const k of Object.keys(receivedArgs)) {
+                      for (const k of Object.keys(
+                        receivedArgs
+                      )) {
                         if (Object.prototype.hasOwnProperty.call(
                           expectedParamsAllowed,
                           k
@@ -1335,6 +1114,274 @@ var bfclParallelMultipleBenchmark = createBfclBenchmark(
   "BFCL_v3_parallel_multiple.json",
   "BFCL_v3_parallel_multiple_possible_answer.json"
 );
+
+// src/benchmarks/json-generation.ts
+var import_ai2 = require("ai");
+var import_ajv = __toESM(require("ajv"), 1);
+var import_fs3 = require("fs");
+var import_path3 = __toESM(require("path"), 1);
+function extractFirstJsonBlock(text) {
+  try {
+    return JSON.parse(text);
+  } catch {
+  }
+  const fenceMatch = text.match(/```json\s*([\s\S]*?)```/i) || text.match(/```\s*([\s\S]*?)```/i);
+  if (fenceMatch) {
+    const inner = fenceMatch[1].trim();
+    try {
+      return JSON.parse(inner);
+    } catch {
+    }
+  }
+  const startIdxObj = text.indexOf("{");
+  const startIdxArr = text.indexOf("[");
+  const start = [startIdxObj, startIdxArr].filter((i) => i >= 0).sort((a, b) => a - b)[0];
+  if (start === void 0) return void 0;
+  const open = text[start] === "{" ? "{" : "[";
+  const close = open === "{" ? "}" : "]";
+  let depth = 0;
+  for (let i = start; i < text.length; i++) {
+    const ch = text[i];
+    if (ch === open) depth++;
+    else if (ch === close) depth--;
+    if (depth === 0) {
+      const candidate = text.slice(start, i + 1);
+      try {
+        return JSON.parse(candidate);
+      } catch {
+      }
+      break;
+    }
+  }
+  return void 0;
+}
+function subsetMatch(expected, actual) {
+  if (expected === null || typeof expected !== "object") {
+    return expected === actual;
+  }
+  if (Array.isArray(expected)) {
+    if (!Array.isArray(actual)) return false;
+    for (let i = 0; i < expected.length; i++) {
+      if (!subsetMatch(expected[i], actual[i])) return false;
+    }
+    return true;
+  }
+  if (actual === null || typeof actual !== "object") return false;
+  const eObj = expected;
+  const aObj = actual;
+  for (const key of Object.keys(eObj)) {
+    if (!subsetMatch(eObj[key], aObj[key])) return false;
+  }
+  return true;
+}
+var jsonGenerationBenchmark = {
+  name: "json-generation",
+  version: "2.1.0",
+  description: "Evaluates schema-compliant JSON generation from natural language using JSON Schema prompts.",
+  async run(model, config) {
+    const logs = [];
+    const ajv = new import_ajv.default({ allErrors: true, strict: false });
+    let schemaValidCount = 0;
+    let valueMatchCount = 0;
+    let correctCount = 0;
+    let tests = [];
+    const expectedMap = /* @__PURE__ */ new Map();
+    try {
+      const dataDir = resolveDataDir();
+      const testsJsonl = await import_fs3.promises.readFile(
+        import_path3.default.join(dataDir, "json_generation_tests.jsonl"),
+        "utf-8"
+      );
+      const expectedJsonl = await import_fs3.promises.readFile(
+        import_path3.default.join(dataDir, "json_generation_expected.jsonl"),
+        "utf-8"
+      );
+      tests = testsJsonl.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
+      const expecteds = expectedJsonl.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
+      for (const r of expecteds) expectedMap.set(r.id, r);
+    } catch (e) {
+      const msg = e instanceof Error ? e.message : String(e);
+      return {
+        score: 0,
+        success: false,
+        metrics: {},
+        logs: [`[FATAL] Failed to load json-generation datasets: ${msg}`],
+        error: e
+      };
+    }
+    for (const tc of tests) {
+      try {
+        const schemaStr = JSON.stringify(tc.schema, null, 2);
+        const messages = [
+          {
+            role: "system",
+            content: "You must output only a single JSON document that strictly conforms to the given JSON Schema. Do not include any extra text or code fences."
+          },
+          {
+            role: "user",
+            content: [
+              "Generate a JSON object that reflects the following facts.",
+              "JSON Schema:",
+              schemaStr,
+              "Facts:",
+              tc.promptFacts,
+              "Output must be a single JSON only, with no additional text."
+            ].join("\n\n")
+          }
+        ];
+        const temp = config?.temperature;
+        const temperature = typeof temp === "number" ? temp : void 0;
+        const { text } = await (0, import_ai2.generateText)({
+          model,
+          messages,
+          ...temperature !== void 0 ? { temperature } : {}
+        });
+        let parsed;
+        try {
+          parsed = extractFirstJsonBlock(text);
+        } catch {
+        }
+        if (parsed === void 0) {
+          logs.push(`[FAIL] ${tc.id}: Unable to parse JSON from model output.`);
+          continue;
+        }
+        const validate = ajv.compile(tc.schema);
+        const valid = validate(parsed);
+        if (valid) schemaValidCount++;
+        else
+          logs.push(
+            `[INFO] ${tc.id}: Schema validation errors: ${(validate.errors || []).map((e) => `${e.instancePath} ${e.message}`).join(", ") || "unknown"}`
+          );
+        const expectedRec = expectedMap.get(tc.id);
+        if (!expectedRec) {
+          logs.push(
+            `[WARN] ${tc.id}: No expected record found. Skipping value match.`
+          );
+        }
+        const valuesOk = expectedRec ? subsetMatch(expectedRec.expected, parsed) : false;
+        if (valuesOk) valueMatchCount++;
+        if (valid && valuesOk) {
+          correctCount++;
+          logs.push(`[PASS] ${tc.id}`);
+        } else {
+          logs.push(
+            `[FAIL] ${tc.id}: schemaValid=${valid}, valuesOk=${valuesOk}. Output=${JSON.stringify(
+              parsed
+            )}`
+          );
+        }
+      } catch (e) {
+        const msg = e instanceof Error ? e.message : String(e);
+        logs.push(`[ERROR] ${tc.id}: ${msg}`);
+      }
+    }
+    const total = tests.length;
+    const score = correctCount / total;
+    return {
+      score,
+      success: score >= 0.8,
+      metrics: {
+        total_cases: total,
+        correct_count: correctCount,
+        schema_valid_count: schemaValidCount,
+        value_match_count: valueMatchCount,
+        accuracy: score
+      },
+      logs
+    };
+  }
+};
+var jsonGenerationSchemaOnlyBenchmark = {
+  name: "json-generation-schema-only",
+  version: "1.0.1",
+  description: "Evaluates whether model outputs strictly conform to the provided JSON Schema (structure only).",
+  async run(model, config) {
+    const logs = [];
+    const ajv = new import_ajv.default({ allErrors: true, strict: false });
+    let tests = [];
+    try {
+      const dataDir = resolveDataDir();
+      const testsJsonl = await import_fs3.promises.readFile(
+        import_path3.default.join(dataDir, "json_generation_tests.jsonl"),
+        "utf-8"
+      );
+      tests = testsJsonl.split(/\r?\n/).filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
+    } catch (e) {
+      const msg = e instanceof Error ? e.message : String(e);
+      return {
+        score: 0,
+        success: false,
+        metrics: {},
+        logs: [`[FATAL] Failed to load schema-only tests: ${msg}`],
+        error: e
+      };
+    }
+    let schemaValidCount = 0;
+    for (const tc of tests) {
+      try {
+        const schemaStr = JSON.stringify(tc.schema, null, 2);
+        const messages = [
+          {
+            role: "system",
+            content: "You must output only a single JSON document that strictly conforms to the given JSON Schema. Do not include any extra text or code fences."
+          },
+          {
+            role: "user",
+            content: [
+              "Generate a JSON object that reflects the following facts.",
+              "JSON Schema:",
+              schemaStr,
+              "Facts:",
+              tc.promptFacts,
+              "Output must be a single JSON only, with no additional text."
+            ].join("\n\n")
+          }
+        ];
+        const temp = config?.temperature;
+        const temperature = typeof temp === "number" ? temp : void 0;
+        const { text } = await (0, import_ai2.generateText)({
+          model,
+          messages,
+          ...temperature !== void 0 ? { temperature } : {}
+        });
+        let parsed;
+        try {
+          parsed = extractFirstJsonBlock(text);
+        } catch {
+        }
+        if (parsed === void 0) {
+          logs.push(`[FAIL] ${tc.id}: Could not parse JSON from model output.`);
+          continue;
+        }
+        const validate = ajv.compile(tc.schema);
+        const valid = validate(parsed);
+        if (valid) {
+          schemaValidCount++;
+          logs.push(`[PASS] ${tc.id}`);
+        } else {
+          logs.push(
+            `[FAIL] ${tc.id}: Schema validation errors: ${(validate.errors || []).map((e) => `${e.instancePath} ${e.message}`).join(", ") || "unknown"}`
+          );
+        }
+      } catch (e) {
+        const msg = e instanceof Error ? e.message : String(e);
+        logs.push(`[ERROR] ${tc.id}: ${msg}`);
+      }
+    }
+    const total = tests.length;
+    const score = total > 0 ? schemaValidCount / total : 0;
+    return {
+      score,
+      success: score >= 0.8,
+      metrics: {
+        total_cases: total,
+        schema_valid_count: schemaValidCount,
+        accuracy: score
+      },
+      logs
+    };
+  }
+};
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   bfclMultipleBenchmark,