poquad

src/about.py

@@ -40,7 +40,9 @@ class Tasks(Enum):
     task25 = Task("polish_eq_bench_first_turn", "first_eqbench,none", "eq_bench_first_turn", "generate_until", 0.0)
     task26 = Task("polish_eq_bench", "average_eqbench,none", "eq_bench", "generate_until", 0.0)
     task20 = Task("polish_poleval2018_task3_test_10k", "word_perplexity,none", "poleval2018_task3_test_10k", "other")
-    # task27 = Task("polish_eq_bench", "revised_eqbench,none", "eq_bench_revised", "other", 0.0)
+    task27 = Task("polish_poquad_reranking", "acc,none", "poquad_reranking", "other", 0.0)
+    task28 = Task("polish_abstractive_poquad_rag", "levenshtein,none", "abstractive_poquad_rag", "other", 0.0)
+    task29 = Task("polish_abstractive_poquad_open_book", "levenshtein,none", "abstractive_poquad_open_book", "other", 0.0)
 
 
 g_tasks = [task.value.benchmark for task in Tasks if task.value.type == "generate_until"]

src/display/utils.py

@@ -37,7 +37,8 @@ auto_eval_column_dict.append(["average_mc", ColumnContent, ColumnContent("Avg mc
 auto_eval_column_dict.append(["average_rag", ColumnContent, ColumnContent("Avg RAG", "number", True)])
 
 for task in Tasks:
-    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
+    show = task.name not in ['polish_poquad_reranking','polish_abstractive_poquad_rag','polish_abstractive_poquad_open_book']
+    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", show)])
 # Model information
 
 auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])