Fix wrong GED. Load all dataset at startup

quiz.py

@@ -42,6 +42,15 @@ BENCHMARKS = {
     },
 }
 
+DATASETS = {
+    dataset_name: load_dataset(
+        BENCHMARKS[dataset_name]["path"],
+        name=BENCHMARKS[dataset_name].get("config_name"),
+        split=BENCHMARKS[dataset_name].get("split", "train"),
+    )
+    for dataset_name in BENCHMARKS 
+}
+
 
 # Dataset specific preprocessing and standardization
 def winogrande_preprocessing(sample):
@@ -65,7 +74,7 @@ def icelandic_sentence_gec_preprocessing(sample):
         f"Inniheldur eftirfarandi málsgrein villu?<p style='margin-left: 25px;'><i>{sample['sentence']}</i></p>"
     )
     new_sample["options"] = "Villa", "Engin villa"
-    new_sample["answer"] = "Engin villa" if sample["correct"] else "Villa"
+    new_sample["answer"] = "Engin villa" if sample["correct"] == "false" else "Villa"
     new_sample["instruction"] = "Valkostir"
     return new_sample
 
@@ -161,12 +170,9 @@ class BenchmarkQuiz:
         return self.state
 
     def load_benchmark(self, benchmark_name: str) -> List[Dict[str, Any]]:
-        dataset = load_dataset(
-            BENCHMARKS[benchmark_name]["path"],
-            name=BENCHMARKS[benchmark_name].get("config_name"),
-            split=BENCHMARKS[benchmark_name].get("split", "train"),
-        )
-        samples = random.sample(list(dataset), 5)
+        dataset = DATASETS[benchmark_name]
+        random_indices = random.sample(range(len(dataset)), 5)
+        samples = dataset.select(random_indices)
         if benchmark_name == "icelandic-winogrande":
             samples = [winogrande_preprocessing(sample) for sample in samples]
         elif benchmark_name == "grammatical-error-detection":