wisent 0.7.379__py3-none-any.whl → 0.7.701__py3-none-any.whl

wisent/examples/scripts/results/test_xnli_evaluation.json

@@ -1,51 +0,0 @@
-{
-  "task_name": "xnli",
-  "model_name": "mock",
-  "evaluator_name": "log_likelihoods",
-  "num_pairs": 2,
-  "all_correct": true,
-  "pairs": [
-    {
-      "pair_id": 0,
-      "prompt": "Decide the relationship of the hypothesis '\u0411\u0440\u043e\u043a \u043d\u0435 \u0437\u0430\u0449\u0438\u0442\u0430\u0432\u0430 \u0425\u0438\u043b\u044a\u0440\u0438.' to the premise '\u0412\u0442\u043e\u0440\u043e\u0442\u043e \u043d\u0438\u0432\u043e \u043d\u0430...",
-      "positive_response": "contradiction",
-      "negative_response": "entailment",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'contradiction' (log_prob=-0.500), Expected: 'contradiction'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'contradiction' (log_prob=-0.500), Expected: 'entailment'"
-      },
-      "both_correct": true
-    },
-    {
-      "pair_id": 1,
-      "prompt": "Decide the relationship of the hypothesis 'Sabol hasimami kwa hayo mashimo yote.' to the premise 'Sa...",
-      "positive_response": "neutral",
-      "negative_response": "contradiction",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'neutral' (log_prob=-0.500), Expected: 'neutral'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'neutral' (log_prob=-0.500), Expected: 'contradiction'"
-      },
-      "both_correct": true
-    }
-  ]
-}

wisent/examples/scripts/results/test_xnli_pairs.json

@@ -1,14 +0,0 @@
-[
-  {
-    "pair_id": 0,
-    "prompt": "Decide the relationship of the hypothesis '\u0411\u0440\u043e\u043a \u043d\u0435 \u0437\u0430\u0449\u0438\u0442\u0430\u0432\u0430 \u0425\u0438\u043b\u044a\u0440\u0438.' to the premise '\u0412\u0442\u043e\u0440\u043e\u0442\u043e \u043d\u0438\u0432\u043e \u043d\u0430 \u043b\u0438\u0446\u0435\u043c\u0435\u0440\u0438\u0435 \u0435, \u0447\u0435 \u0411\u0440\u043e\u043a \u0437\u0430\u0449\u0438\u0442\u0430\u0432\u0430 \u0425\u0438\u043b\u0430\u0440\u0438 \u0441\u0430\u043c\u043e, \u0437\u0430 \u0434\u0430 \u0438\u0437\u0434\u0438\u0433\u043d\u0435 \u0441\u0432\u043e\u044f \u0441\u043e\u0431\u0441\u0442\u0432\u0435\u043d \u0441\u043a\u0430\u043d\u0434\u0430\u043b.\nA. entailment\nB. contradiction",
-    "positive_response": "contradiction",
-    "negative_response": "entailment"
-  },
-  {
-    "pair_id": 1,
-    "prompt": "Decide the relationship of the hypothesis 'Sabol hasimami kwa hayo mashimo yote.' to the premise 'Sabol alisema anastahili kuacha shimo la mara kwa mara, lakini hata hilo linapangwa kwa urahisi.\nA. contradiction\nB. neutral",
-    "positive_response": "neutral",
-    "negative_response": "contradiction"
-  }
-]

wisent/examples/scripts/results/test_xquad_evaluation.json

@@ -1,51 +0,0 @@
-{
-  "task_name": "xquad",
-  "model_name": "mock",
-  "evaluator_name": "log_likelihoods",
-  "num_pairs": 2,
-  "all_correct": true,
-  "pairs": [
-    {
-      "pair_id": 0,
-      "prompt": "Context: Es gibt einige verbreitete Irrglauben \u00fcber die \u00e4u\u00dferen und inneren Chloroplastenmembranen. ...",
-      "positive_response": "das Produkt der Zellmembran des Wirts ist, die sich entfaltet, um ein das anzestrale Zyanobakterium umgebendes Vesikel zu bilden",
-      "negative_response": "Es gibt einige",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'das Produkt der Zellmembran des Wirts ist, die sich entfaltet, um ein das anzestrale Zyanobakterium umgebendes Vesikel zu bilden' (log_prob=-0.500), Expected: 'das Produkt der Zellmembran des Wirts ist, die sich entfaltet, um ein das anzestrale Zyanobakterium umgebendes Vesikel zu bilden'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'das Produkt der Zellmembran des Wirts ist, die sich entfaltet, um ein das anzestrale Zyanobakterium umgebendes Vesikel zu bilden' (log_prob=-0.500), Expected: 'Es gibt einige'"
-      },
-      "both_correct": true
-    },
-    {
-      "pair_id": 1,
-      "prompt": "Context: Incluso antes de que Washington regresara, Dinwiddie hab\u00eda enviado una compa\u00f1\u00eda de 40 hombr...",
-      "positive_response": "1754",
-      "negative_response": "Incluso antes de",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '1754' (log_prob=-0.500), Expected: '1754'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '1754' (log_prob=-0.500), Expected: 'Incluso antes de'"
-      },
-      "both_correct": true
-    }
-  ]
-}

wisent/examples/scripts/results/test_xquad_pairs.json

@@ -1,14 +0,0 @@
-[
-  {
-    "pair_id": 0,
-    "prompt": "Context: Es gibt einige verbreitete Irrglauben \u00fcber die \u00e4u\u00dferen und inneren Chloroplastenmembranen. Die Tatsache, dass Chloroplasten von einer Doppelmembran umgeben sind, wird oft als Beweis daf\u00fcr angef\u00fchrt, dass sie die Vorfahren endosymbiotischer Cyanobakterien sind. Dies wird oft so verstanden, dass die \u00e4u\u00dfere Chloroplastenmembran das Produkt der Zellmembran des Wirts ist, die sich entfaltet, um ein das anzestrale Zyanobakterium umgebendes Vesikel zu bilden \u2013 was unwahr ist. Beide Chloroplasten-Membranen sind den urspr\u00fcnglichen Doppelmembranen des Cyanobakteriums homolog.\n\nQuestion: Was wird f\u00e4lschlicherweise von der \u00e4u\u00dferen Chloroplasten-Membran angenommen?\n\nAnswer:",
-    "positive_response": "das Produkt der Zellmembran des Wirts ist, die sich entfaltet, um ein das anzestrale Zyanobakterium umgebendes Vesikel zu bilden",
-    "negative_response": "Es gibt einige"
-  },
-  {
-    "pair_id": 1,
-    "prompt": "Context: Incluso antes de que Washington regresara, Dinwiddie hab\u00eda enviado una compa\u00f1\u00eda de 40 hombres a las \u00f3rdenes de William Trent hasta ese momento, donde en los primeros meses de 1754 comenzaron la construcci\u00f3n de un peque\u00f1o fuerte empedrado. El Gobernador Duquesne envi\u00f3 fuerzas francesas adicionales bajo el mando de Claude-Pierre Pecaudy de Contrec\u0153ur para relevar a Saint-Pierre durante el mismo per\u00edodo, y Contrec\u0153ur dirigi\u00f3 a 500 hombres al sur de Fort Venango el 5 de abril de 1754. Cuando estas fuerzas llegaron al fuerte el 16 de abril, Contrec\u0153ur permiti\u00f3 generosamente que la peque\u00f1a compa\u00f1\u00eda de Trent se retirara. Compr\u00f3 sus herramientas de construcci\u00f3n para seguir construyendo lo que se convirti\u00f3 en Fort Duquesne.\n\nQuestion: \u00bfCu\u00e1ndo comenzaron los brit\u00e1nicos a construir el fuerte a las \u00f3rdenes de William Trent?\n\nAnswer:",
-    "positive_response": "1754",
-    "negative_response": "Incluso antes de"
-  }
-]

wisent/examples/scripts/results/test_xstorycloze_evaluation.json

@@ -1,51 +0,0 @@
-{
-  "task_name": "xstorycloze",
-  "model_name": "mock",
-  "evaluator_name": "log_likelihoods",
-  "num_pairs": 2,
-  "all_correct": true,
-  "pairs": [
-    {
-      "pair_id": 0,
-      "prompt": "\u0c39\u0c48 \u0c38\u0c4d\u0c15\u0c42\u0c32\u0c4d \u0c35\u0c3f\u0c26\u0c4d\u0c2f\u0c3e\u0c30\u0c4d\u0c25\u0c3f \u0c21\u0c46\u0c02\u0c1f\u0c3f\u0c38\u0c4d\u0c1f\u0c4d \u0c26\u0c17\u0c4d\u0c17\u0c30\u0c15\u0c41 \u0c35\u0c46\u0c33\u0c4d\u0c32\u0c3f\u0c02\u0c26\u0c3f. \u0c06\u0c2e\u0c46 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c41 \u0c35\u0c47\u0c38\u0c41\u0c15\u0c4b\u0c35\u0c3e\u0c32\u0c28\u0c3f \u0c1a\u0c46\u0c2a\u0c4d\u0c2a\u0c3e\u0c30\u0c41. \u0c06\u0c2e\u0c46\u0c15\u0c41 \u0c05\u0c26\u0c3f \u0c28\u0c1a\u0c4d\u0c1a...",
-      "positive_response": "\u0c06\u0c2e\u0c46 \u0c2a\u0c33\u0c4d\u0c32\u0c41 \u0c38\u0c4d\u0c1f\u0c4d\u0c30\u0c2f\u0c3f\u0c1f\u0c4d\u200c\u0c17\u0c3e \u0c35\u0c1a\u0c4d\u0c1a\u0c3e\u0c15 \u0c06\u0c2e\u0c46 \u0c38\u0c02\u0c24\u0c4b\u0c37\u0c3f\u0c02\u0c1a\u0c3f\u0c02\u0c26\u0c3f.",
-      "negative_response": "\u0c2a\u0c4d\u0c30\u0c24\u0c3f \u0c38\u0c02\u0c26\u0c30\u0c4d\u0c36\u0c28\u0c32\u0c4b \u0c24\u0c28 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c28\u0c41 \u0c1f\u0c2f\u0c3f\u0c1f\u0c4d \u0c1a\u0c47\u0c2f\u0c3f\u0c02\u0c1a\u0c41\u0c15\u0c4b\u0c35\u0c21\u0c3e\u0c28\u0c3f\u0c15\u0c3f \u0c06\u0c2e\u0c46 \u0c0e\u0c26\u0c41\u0c30\u0c41 \u0c1a\u0c42\u0c38\u0c4d\u0c24\u0c4b\u0c02\u0c26\u0c3f.",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '\u0c06\u0c2e\u0c46 \u0c2a\u0c33\u0c4d\u0c32\u0c41 \u0c38\u0c4d\u0c1f\u0c4d\u0c30\u0c2f\u0c3f\u0c1f\u0c4d\u200c\u0c17\u0c3e \u0c35\u0c1a\u0c4d\u0c1a\u0c3e\u0c15 \u0c06\u0c2e\u0c46 \u0c38\u0c02\u0c24\u0c4b\u0c37\u0c3f\u0c02\u0c1a\u0c3f\u0c02\u0c26\u0c3f.' (log_prob=-0.500), Expected: '\u0c06\u0c2e\u0c46 \u0c2a\u0c33\u0c4d\u0c32\u0c41 \u0c38\u0c4d\u0c1f\u0c4d\u0c30\u0c2f\u0c3f\u0c1f\u0c4d\u200c\u0c17\u0c3e \u0c35\u0c1a\u0c4d\u0c1a\u0c3e\u0c15 \u0c06\u0c2e\u0c46 \u0c38\u0c02\u0c24\u0c4b\u0c37\u0c3f\u0c02\u0c1a\u0c3f\u0c02\u0c26\u0c3f.'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '\u0c06\u0c2e\u0c46 \u0c2a\u0c33\u0c4d\u0c32\u0c41 \u0c38\u0c4d\u0c1f\u0c4d\u0c30\u0c2f\u0c3f\u0c1f\u0c4d\u200c\u0c17\u0c3e \u0c35\u0c1a\u0c4d\u0c1a\u0c3e\u0c15 \u0c06\u0c2e\u0c46 \u0c38\u0c02\u0c24\u0c4b\u0c37\u0c3f\u0c02\u0c1a\u0c3f\u0c02\u0c26\u0c3f.' (log_prob=-0.500), Expected: '\u0c2a\u0c4d\u0c30\u0c24\u0c3f \u0c38\u0c02\u0c26\u0c30\u0c4d\u0c36\u0c28\u0c32\u0c4b \u0c24\u0c28 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c28\u0c41 \u0c1f\u0c2f\u0c3f\u0c1f\u0c4d \u0c1a\u0c47\u0c2f\u0c3f\u0c02\u0c1a\u0c41\u0c15\u0c4b\u0c35\u0c21\u0c3e\u0c28\u0c3f\u0c15\u0c3f \u0c06\u0c2e\u0c46 \u0c0e\u0c26\u0c41\u0c30\u0c41 \u0c1a\u0c42\u0c38\u0c4d\u0c24\u0c4b\u0c02\u0c26\u0c3f.'"
-      },
-      "both_correct": true
-    },
-    {
-      "pair_id": 1,
-      "prompt": "\u4e00\u5929\u65e9\u4e0a\uff0c\u8428\u7c73\u80c3\u75bc\u5f97\u5389\u5bb3\u3002 \u4f46\u5979\u5f3a\u8feb\u81ea\u5df1\u53bb\u4e0a\u5b66\u3002 \u5f53\u5979\u8d70\u8fdb\u516c\u5171\u6c7d\u8f66\u65f6\uff0c\u5979\u60f3\u5f80\u56de\u8d70\u3002 \u4f46\u5979\u89c9\u5f97\u73b0\u5728\u6539\u53d8\u4e3b\u610f\u592a\u665a\u4e86\u3002\n \nA. \u5979\u5750\u5728\u8def\u8fb9\uff0c\u770b\u7740\u516c\u5171\u6c7d\u8f66\u4ece\u8eab\u8fb9\u9a76\u8fc7\u3002\nB. \u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002...",
-      "positive_response": "\u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002",
-      "negative_response": "\u5979\u5750\u5728\u8def\u8fb9\uff0c\u770b\u7740\u516c\u5171\u6c7d\u8f66\u4ece\u8eab\u8fb9\u9a76\u8fc7\u3002",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '\u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002' (log_prob=-0.500), Expected: '\u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '\u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002' (log_prob=-0.500), Expected: '\u5979\u5750\u5728\u8def\u8fb9\uff0c\u770b\u7740\u516c\u5171\u6c7d\u8f66\u4ece\u8eab\u8fb9\u9a76\u8fc7\u3002'"
-      },
-      "both_correct": true
-    }
-  ]
-}

wisent/examples/scripts/results/test_xstorycloze_pairs.json

@@ -1,14 +0,0 @@
-[
-  {
-    "pair_id": 0,
-    "prompt": "\u0c39\u0c48 \u0c38\u0c4d\u0c15\u0c42\u0c32\u0c4d \u0c35\u0c3f\u0c26\u0c4d\u0c2f\u0c3e\u0c30\u0c4d\u0c25\u0c3f \u0c21\u0c46\u0c02\u0c1f\u0c3f\u0c38\u0c4d\u0c1f\u0c4d \u0c26\u0c17\u0c4d\u0c17\u0c30\u0c15\u0c41 \u0c35\u0c46\u0c33\u0c4d\u0c32\u0c3f\u0c02\u0c26\u0c3f. \u0c06\u0c2e\u0c46 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c41 \u0c35\u0c47\u0c38\u0c41\u0c15\u0c4b\u0c35\u0c3e\u0c32\u0c28\u0c3f \u0c1a\u0c46\u0c2a\u0c4d\u0c2a\u0c3e\u0c30\u0c41. \u0c06\u0c2e\u0c46\u0c15\u0c41 \u0c05\u0c26\u0c3f \u0c28\u0c1a\u0c4d\u0c1a\u0c32\u0c47\u0c26\u0c41. \u0c06\u0c2e\u0c46 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c41 \u0c35\u0c47\u0c38\u0c41\u0c15\u0c41\u0c02\u0c26\u0c3f, \u0c1a\u0c3e\u0c32\u0c3e \u0c28\u0c4a\u0c2a\u0c4d\u0c2a\u0c3f \u0c09\u0c02\u0c26\u0c3f.\n \nA. \u0c2a\u0c4d\u0c30\u0c24\u0c3f \u0c38\u0c02\u0c26\u0c30\u0c4d\u0c36\u0c28\u0c32\u0c4b \u0c24\u0c28 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c28\u0c41 \u0c1f\u0c2f\u0c3f\u0c1f\u0c4d \u0c1a\u0c47\u0c2f\u0c3f\u0c02\u0c1a\u0c41\u0c15\u0c4b\u0c35\u0c21\u0c3e\u0c28\u0c3f\u0c15\u0c3f \u0c06\u0c2e\u0c46 \u0c0e\u0c26\u0c41\u0c30\u0c41 \u0c1a\u0c42\u0c38\u0c4d\u0c24\u0c4b\u0c02\u0c26\u0c3f.\nB. \u0c06\u0c2e\u0c46 \u0c2a\u0c33\u0c4d\u0c32\u0c41 \u0c38\u0c4d\u0c1f\u0c4d\u0c30\u0c2f\u0c3f\u0c1f\u0c4d\u200c\u0c17\u0c3e \u0c35\u0c1a\u0c4d\u0c1a\u0c3e\u0c15 \u0c06\u0c2e\u0c46 \u0c38\u0c02\u0c24\u0c4b\u0c37\u0c3f\u0c02\u0c1a\u0c3f\u0c02\u0c26\u0c3f.",
-    "positive_response": "\u0c06\u0c2e\u0c46 \u0c2a\u0c33\u0c4d\u0c32\u0c41 \u0c38\u0c4d\u0c1f\u0c4d\u0c30\u0c2f\u0c3f\u0c1f\u0c4d\u200c\u0c17\u0c3e \u0c35\u0c1a\u0c4d\u0c1a\u0c3e\u0c15 \u0c06\u0c2e\u0c46 \u0c38\u0c02\u0c24\u0c4b\u0c37\u0c3f\u0c02\u0c1a\u0c3f\u0c02\u0c26\u0c3f.",
-    "negative_response": "\u0c2a\u0c4d\u0c30\u0c24\u0c3f \u0c38\u0c02\u0c26\u0c30\u0c4d\u0c36\u0c28\u0c32\u0c4b \u0c24\u0c28 \u0c2c\u0c4d\u0c30\u0c47\u0c38\u0c4d\u200c\u0c32\u0c28\u0c41 \u0c1f\u0c2f\u0c3f\u0c1f\u0c4d \u0c1a\u0c47\u0c2f\u0c3f\u0c02\u0c1a\u0c41\u0c15\u0c4b\u0c35\u0c21\u0c3e\u0c28\u0c3f\u0c15\u0c3f \u0c06\u0c2e\u0c46 \u0c0e\u0c26\u0c41\u0c30\u0c41 \u0c1a\u0c42\u0c38\u0c4d\u0c24\u0c4b\u0c02\u0c26\u0c3f."
-  },
-  {
-    "pair_id": 1,
-    "prompt": "\u4e00\u5929\u65e9\u4e0a\uff0c\u8428\u7c73\u80c3\u75bc\u5f97\u5389\u5bb3\u3002 \u4f46\u5979\u5f3a\u8feb\u81ea\u5df1\u53bb\u4e0a\u5b66\u3002 \u5f53\u5979\u8d70\u8fdb\u516c\u5171\u6c7d\u8f66\u65f6\uff0c\u5979\u60f3\u5f80\u56de\u8d70\u3002 \u4f46\u5979\u89c9\u5f97\u73b0\u5728\u6539\u53d8\u4e3b\u610f\u592a\u665a\u4e86\u3002\n \nA. \u5979\u5750\u5728\u8def\u8fb9\uff0c\u770b\u7740\u516c\u5171\u6c7d\u8f66\u4ece\u8eab\u8fb9\u9a76\u8fc7\u3002\nB. \u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002",
-    "positive_response": "\u5979\u53bb\u4e0a\u5b66\uff0c\u53d7\u7740\u75db\u82e6\u3002",
-    "negative_response": "\u5979\u5750\u5728\u8def\u8fb9\uff0c\u770b\u7740\u516c\u5171\u6c7d\u8f66\u4ece\u8eab\u8fb9\u9a76\u8fc7\u3002"
-  }
-]

wisent/examples/scripts/results/test_xsum_evaluation.json

@@ -1,30 +0,0 @@
-{
-  "task_name": "xsum",
-  "model_name": "mock",
-  "evaluator_name": "log_likelihoods",
-  "num_pairs": 1,
-  "all_correct": true,
-  "pairs": [
-    {
-      "pair_id": 0,
-      "prompt": "Summarize the following document: The ex-Reading defender denied fraudulent trading charges relating...",
-      "positive_response": "Former Premier League footballer Sam Sodje has appeared in court alongside three brothers accused of charity fraud.",
-      "negative_response": "Former Premier League footballer Sam Sodje has appeared",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'Former Premier League footballer Sam Sodje has appeared in court alongside three brothers accused of charity fraud.' (log_prob=-0.500), Expected: 'Former Premier League footballer Sam Sodje has appeared in court alongside three brothers accused of charity fraud.'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'Former Premier League footballer Sam Sodje has appeared in court alongside three brothers accused of charity fraud.' (log_prob=-0.500), Expected: 'Former Premier League footballer Sam Sodje has appeared'"
-      },
-      "both_correct": true
-    }
-  ]
-}

wisent/examples/scripts/results/test_xsum_pairs.json

@@ -1,8 +0,0 @@
-[
-  {
-    "pair_id": 0,
-    "prompt": "Summarize the following document: The ex-Reading defender denied fraudulent trading charges relating to the Sodje Sports Foundation - a charity to raise money for Nigerian sport.\nMr Sodje, 37, is jointly charged with elder brothers Efe, 44, Bright, 50 and Stephen, 42.\nAppearing at the Old Bailey earlier, all four denied the offence.\nThe charge relates to offences which allegedly took place between 2008 and 2014.\nSam, from Kent, Efe and Bright, of Greater Manchester, and Stephen, from Bexley, are due to stand trial in July.\nThey were all released on bail..",
-    "positive_response": "Former Premier League footballer Sam Sodje has appeared in court alongside three brothers accused of charity fraud.",
-    "negative_response": "Former Premier League footballer Sam Sodje has appeared"
-  }
-]

wisent/examples/scripts/results/test_xwinograd_evaluation.json

@@ -1,51 +0,0 @@
-{
-  "task_name": "xwinograd",
-  "model_name": "mock",
-  "evaluator_name": "log_likelihoods",
-  "num_pairs": 2,
-  "all_correct": true,
-  "pairs": [
-    {
-      "pair_id": 0,
-      "prompt": "Fill in the blank: The police arrested the rioters because _ were preventing trouble.\nA. the rioters...",
-      "positive_response": "The police",
-      "negative_response": "the rioters",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'The police' (log_prob=-0.500), Expected: 'The police'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'The police' (log_prob=-0.500), Expected: 'the rioters'"
-      },
-      "both_correct": true
-    },
-    {
-      "pair_id": 1,
-      "prompt": "Fill in the blank: \u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8\u306f\u4fa1\u683c\u3067\u6d88\u8cbb\u8005\u3092\u62b1\u304d\u8fbc\u307e\u306a\u3051\u308c\u3070\u306a\u3089\u306a\u3044\u3002\u3055\u3082\u306a\u3051\u308c\u3070_\u306f\u5012\u7523\u3059\u308b\u3053\u3068\u306b\u306a\u308a\u305d\u3046\u3060\u3002\nA. \u6d88\u8cbb\u8005\nB. \u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8...",
-      "positive_response": "\u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8",
-      "negative_response": "\u6d88\u8cbb\u8005",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '\u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8' (log_prob=-0.500), Expected: '\u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: '\u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8' (log_prob=-0.500), Expected: '\u6d88\u8cbb\u8005'"
-      },
-      "both_correct": true
-    }
-  ]
-}

wisent/examples/scripts/results/test_xwinograd_pairs.json

@@ -1,14 +0,0 @@
-[
-  {
-    "pair_id": 0,
-    "prompt": "Fill in the blank: The police arrested the rioters because _ were preventing trouble.\nA. the rioters\nB. The police",
-    "positive_response": "The police",
-    "negative_response": "the rioters"
-  },
-  {
-    "pair_id": 1,
-    "prompt": "Fill in the blank: \u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8\u306f\u4fa1\u683c\u3067\u6d88\u8cbb\u8005\u3092\u62b1\u304d\u8fbc\u307e\u306a\u3051\u308c\u3070\u306a\u3089\u306a\u3044\u3002\u3055\u3082\u306a\u3051\u308c\u3070_\u306f\u5012\u7523\u3059\u308b\u3053\u3068\u306b\u306a\u308a\u305d\u3046\u3060\u3002\nA. \u6d88\u8cbb\u8005\nB. \u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8",
-    "positive_response": "\u30a8\u30f3\u30bf\u30fc\u30c6\u30a4\u30f3\u30de\u30fc\u30c8",
-    "negative_response": "\u6d88\u8cbb\u8005"
-  }
-]

wisent/examples/scripts/results/test_yahoo_answers_topics_evaluation.json

@@ -1,30 +0,0 @@
-{
-  "task_name": "yahoo_answers_topics",
-  "model_name": "mock",
-  "evaluator_name": "log_likelihoods",
-  "num_pairs": 1,
-  "all_correct": true,
-  "pairs": [
-    {
-      "pair_id": 0,
-      "prompt": "Classify the Topic of the following Text to one of these options: Society & Culture, Science & Mathe...",
-      "positive_response": "Sports",
-      "negative_response": "Society & Culture",
-      "positive_evaluation": {
-        "ground_truth": "TRUTHFUL",
-        "confidence": 1.0,
-        "expected": "TRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'Sports' (log_prob=-0.500), Expected: 'Sports'"
-      },
-      "negative_evaluation": {
-        "ground_truth": "UNTRUTHFUL",
-        "confidence": 0.0,
-        "expected": "UNTRUTHFUL",
-        "correct": true,
-        "details": "Predicted: 'Sports' (log_prob=-0.500), Expected: 'Society & Culture'"
-      },
-      "both_correct": true
-    }
-  ]
-}

wisent/examples/scripts/results/test_yahoo_answers_topics_pairs.json

@@ -1,8 +0,0 @@
-[
-  {
-    "pair_id": 0,
-    "prompt": "Classify the Topic of the following Text to one of these options: Society & Culture, Science & Mathematics, Health, Education & Reference, Computers & Internet, Sports, Business & Finance, Entertainment & Music, Family & Relationships, Politics & Government.\nText:\nwhere can i volenteer for the 2007 pga championship?  Volunteer information for the 2007 PGA Championship at Southern Hills Country Club in Tulsa, OK is not yet available. It will likely become available soon after the completion of the 2006 Championship at Medinah outside of Chicago Aug. 14-20. Check out the following site for the 2007, which is where the info will be posted as soon as it is available in the event info section of the site.\nTopic:",
-    "positive_response": "Sports",
-    "negative_response": "Society & Culture"
-  }
-]