wisent 0.7.701__py3-none-any.whl → 0.7.1045__py3-none-any.whl

wisent/parameters/lm_eval/broken_in_lm_eval.json

@@ -1,10 +1,187 @@
 [
+  "20_newsgroups",
+  "acp_bench_hard",
+  "acpbench",
+  "african_flores",
+  "afrimgsm",
+  "afrobench_adr",
+  "ag_news",
+  "agentharm",
+  "agieval",
+  "aime2024",
+  "aime2025",
+  "alpaca_eval",
+  "anagrams1",
+  "anagrams2",
+  "aradice",
+  "argument_topic",
+  "assin_entailment",
+  "banking77",
+  "basque_bench",
+  "benchmarks",
+  "bfcl",
+  "bhs",
+  "bhtc",
+  "bigbench",
+  "bigbench_generate_until",
+  "cabbq",
+  "cabreu_abstractive",
+  "chain",
+  "claim_stance_topic",
+  "click",
+  "cluewsc",
+  "cnn_dailymail",
+  "cocoteros_va",
+  "code2text",
   "code_x_glue",
+  "codeforces",
+  "codexglue_code_to_text_go",
+  "codexglue_code_to_text_java",
+  "codexglue_code_to_text_javascript",
+  "codexglue_code_to_text_php",
+  "codexglue_code_to_text_python",
+  "codexglue_code_to_text_ruby",
+  "coedit",
+  "copa_ca",
+  "cycle_letters",
+  "dbpedia_14",
+  "discrim_eval",
+  "doc",
+  "egyhellaswag",
+  "egymmlu",
+  "epec",
   "epec_koref_bin",
+  "esbbq",
+  "escola",
+  "ethos_binary",
+  "evalita-mp",
+  "evalita-sp",
+  "financial_tweets",
+  "flan",
   "flan_held_in",
+  "flores_ca-eu",
+  "gpt3_translation_benchmarks",
+  "harmbench",
+  "hle",
+  "icelandic_winogrande",
+  "ifeval",
+  "instruct_humaneval",
+  "instructhumaneval",
+  "jailbreakbench",
+  "japanese_leaderboard",
+  "jsonschema_bench",
+  "kmmlu",
+  "kmmlu_accounting",
+  "law_stack_exchange",
+  "leaderboard",
+  "ledgar",
+  "libra",
+  "librusec_history",
+  "livemathbench_cnmo_zh",
+  "llama3",
+  "lm_syneval",
+  "long_context_multiq",
+  "longbench",
+  "longbenchv2",
+  "matreshka_names",
+  "mbpp",
+  "mbpp_plus",
+  "mc-taco",
+  "medical_abstracts",
+  "mediqa_qa2019",
+  "medtext",
+  "meqsum",
+  "mgsm_direct_eu",
+  "mimic_repsum",
+  "minerva_math",
+  "mmlu_redux",
+  "mmmlu",
+  "mts_dialog",
+  "multi_swe_bench",
+  "multiblimp",
+  "multilingual",
+  "ncb",
+  "niah_single_1",
+  "norbelebele_p0",
+  "norec_document_p0",
+  "noreval",
+  "noropenbookqa_nno_p0",
+  "norrewrite_instruct",
+  "norsumm_nno_p0",
+  "norsummarize_instruct",
+  "okapi",
+  "okapi/arc_multilingual",
+  "olaph",
+  "olympiadbench",
+  "openbookqa_ca",
+  "or_bench",
+  "parafraseja",
+  "parafrases_gl",
+  "passkey",
+  "penn_treebank",
+  "pile",
+  "pile_10k",
+  "piqa_eu",
+  "portuguese_bench",
+  "ptb",
+  "qnlieu",
+  "quac",
+  "random",
+  "reversed",
+  "ru_2wikimultihopqa",
   "ruler",
+  "scrolls",
+  "sglue",
+  "social_iqa",
+  "sorry_bench",
+  "squad2",
+  "summarization_gl",
+  "super-glue-t5-prompt",
+  "super_glue",
+  "super_glue-wsc-t5-prompt",
+  "superglue",
+  "supergpqa",
+  "swe_bench_multilingual",
+  "sycophancy_eval",
+  "t0",
   "t0_eval",
+  "tatoeba_eng_nno_p0",
+  "teca",
+  "tinyArc",
+  "tinyBenchmarks",
+  "tinyGSM8k",
+  "tinyHellaswag",
+  "tinyMMLU",
+  "tinyTruthfulQA",
+  "tinyWinogrande",
+  "tinyarc",
+  "tinybenchmarks",
+  "tinygsm8k",
+  "tinyhellaswag",
+  "tinymmlu",
+  "tinytruthfulqa",
+  "tinywinogrande",
   "tmlu",
+  "trasnlation_all_flores",
+  "truthfulqa-multi",
+  "truthfulqa_gl_mc1",
+  "turblimp_core",
+  "turkishmmlu",
+  "turkishmmlu_biology",
+  "twenty_newsgroups",
+  "unfair_tos",
+  "unitxt",
+  "unscramble",
+  "vaxx",
   "vaxx_stance",
-  "wiceu"
-]
+  "wiceu",
+  "wikitext103",
+  "wildguard",
+  "wmt16-en-ro",
+  "wmt16-ro-en",
+  "wmt16_en_ro",
+  "wmt16_ro_en",
+  "wmt2014",
+  "xlsum_es",
+  "xnli_gl"
+]

wisent/parameters/lm_eval/category_directions.json

@@ -0,0 +1,137 @@
+{
+  "coding": {
+    "description": "Code generation, understanding, and debugging capabilities",
+    "hypothesized_directions": [
+      "code_correctness",
+      "code_completeness",
+      "algorithmic_thinking",
+      "code_style"
+    ]
+  },
+  "math": {
+    "description": "Mathematical reasoning and computation",
+    "hypothesized_directions": [
+      "numerical_accuracy",
+      "algebraic_reasoning",
+      "problem_decomposition",
+      "mathematical_rigor"
+    ]
+  },
+  "reasoning_logic": {
+    "description": "Logical deduction and multi-step reasoning",
+    "hypothesized_directions": [
+      "deductive_reasoning",
+      "causal_reasoning",
+      "planning",
+      "constraint_satisfaction"
+    ]
+  },
+  "hallucination_factuality": {
+    "description": "Truthfulness and factual accuracy",
+    "hypothesized_directions": [
+      "factual_recall",
+      "uncertainty_awareness",
+      "source_grounding",
+      "confabulation_resistance"
+    ]
+  },
+  "safety_bias": {
+    "description": "Safety, fairness, and bias mitigation",
+    "hypothesized_directions": [
+      "harm_avoidance",
+      "stereotype_resistance",
+      "fairness",
+      "toxicity_avoidance"
+    ]
+  },
+  "multilingual": {
+    "description": "Cross-lingual and language-specific capabilities",
+    "hypothesized_directions": [
+      "language_transfer",
+      "cultural_awareness",
+      "script_handling",
+      "cross_lingual_consistency"
+    ]
+  },
+  "knowledge_qa": {
+    "description": "World knowledge and question answering",
+    "hypothesized_directions": [
+      "factual_knowledge",
+      "knowledge_retrieval",
+      "answer_precision",
+      "domain_expertise"
+    ]
+  },
+  "reading_comprehension": {
+    "description": "Understanding and extracting information from text",
+    "hypothesized_directions": [
+      "information_extraction",
+      "inference_making",
+      "context_tracking",
+      "summarization"
+    ]
+  },
+  "commonsense": {
+    "description": "Everyday reasoning and world understanding",
+    "hypothesized_directions": [
+      "physical_intuition",
+      "social_reasoning",
+      "temporal_reasoning",
+      "spatial_reasoning"
+    ]
+  },
+  "science_medical": {
+    "description": "Scientific and medical domain knowledge",
+    "hypothesized_directions": [
+      "scientific_accuracy",
+      "medical_knowledge",
+      "technical_precision",
+      "evidence_based_reasoning"
+    ]
+  },
+  "instruction_following": {
+    "description": "Following complex instructions and user intent",
+    "hypothesized_directions": [
+      "instruction_adherence",
+      "format_compliance",
+      "constraint_following",
+      "intent_understanding"
+    ]
+  },
+  "tool_use_agents": {
+    "description": "Using tools and acting as an agent",
+    "hypothesized_directions": [
+      "tool_selection",
+      "api_usage",
+      "action_planning",
+      "error_recovery"
+    ]
+  },
+  "language_understanding": {
+    "description": "Core linguistic competence",
+    "hypothesized_directions": [
+      "syntactic_knowledge",
+      "semantic_understanding",
+      "pragmatic_competence",
+      "lexical_knowledge"
+    ]
+  },
+  "translation": {
+    "description": "Cross-lingual translation",
+    "hypothesized_directions": [
+      "translation_accuracy",
+      "fluency",
+      "terminology_handling",
+      "style_preservation"
+    ]
+  },
+  "ethics_values": {
+    "description": "Ethical reasoning and value alignment",
+    "hypothesized_directions": [
+      "moral_reasoning",
+      "value_consistency",
+      "norm_awareness",
+      "ethical_sensitivity"
+    ]
+  }
+}

wisent/parameters/lm_eval/repair_plan.json

@@ -0,0 +1,282 @@
+{
+  "summary": {
+    "total_tested": 321,
+    "ok": 179,
+    "failed": 142,
+    "success_rate": "56%"
+  },
+  "case_sensitive_fix": {
+    "description": "Task exists in lm-eval but our code uses wrong case",
+    "repair": "Fix case in manifest or loader",
+    "tasks": [
+      {"our_name": "AraDiCE_ArabicMMLU_lev", "correct_name": "AraDiCE_ArabicMMLU_lev", "note": "timeout issue, not case"},
+      {"our_name": "aexams_IslamicStudies", "correct_name": "aexams_IslamicStudies", "note": "our code lowercases to aexams_islamicstudies"}
+    ]
+  },
+  "task_name_mappings": {
+    "description": "Task exists in lm-eval but under different name - verified with lm_eval --tasks",
+    "repair": "Update manifest to use correct lm-eval task name",
+    "tasks": [
+      {
+        "our_name": "mc-taco",
+        "correct_name": "mc_taco",
+        "reason": "lm_eval --tasks mc-taco fails, lm_eval --tasks mc_taco works (hyphen vs underscore)"
+      }
+    ]
+  },
+  "lm_eval_bugs": {
+    "description": "Folder exists in lm-eval/tasks but group/task name not registered - bug in lm-eval harness",
+    "repair": "Use one of the existing subtasks, or wait for lm-eval fix, or move to HF_EXTRACTORS",
+    "tasks": [
+      {
+        "our_name": "acpbench",
+        "folder": "lm_eval/tasks/acpbench",
+        "note": "Folder exists but task 'acpbench' not registered. README says task is 'acp_bench'",
+        "available_groups": ["acp_bench", "acp_bench_hard", "acp_bench_hard_with_pddl"],
+        "available_subtasks": ["acp_bool_cot_2shot", "acp_gen_2shot", "acp_mcq_cot_2shot", "acp_app_bool", "etc."]
+      },
+      {
+        "our_name": "afrimgsm",
+        "folder": "lm_eval/tasks/afrimgsm",
+        "note": "Folder README says 'afrimgsm: All afrimgsm tasks' but this group is NOT defined in any YAML. YAMLs define 'afrimgsm-irokobench' etc.",
+        "available_groups": ["afrimgsm-irokobench", "afrimgsm_cot-irokobench", "afrimgsm_tt-irokobench", "afrimgsm_tt_cot-irokobench"],
+        "available_subtasks": ["afrimgsm_tasks_prompt_1", "afrimgsm_amh_prompt_1", "etc."]
+      },
+      {
+        "our_name": "afrimmlu",
+        "folder": "lm_eval/tasks/afrimmlu",
+        "note": "Same issue as afrimgsm - folder exists, group 'afrimmlu' not registered",
+        "available_groups": ["afrimmlu-irokobench"]
+      },
+      {
+        "our_name": "llama3",
+        "folder": "lm_eval/tasks/llama3",
+        "note": "Folder exists but no 'llama3' group. Contains MMLU variants for Llama evaluation",
+        "available_groups": ["mmlu_llama", "mmlu_cot_llama", "mmlu_de_llama", "mmlu_es_llama", "mmlu_fr_llama", "etc."]
+      },
+      {
+        "our_name": "mmmlu",
+        "folder": "lm_eval/tasks/okapi/mmlu_multilingual",
+        "note": "No 'mmmlu' task. Multilingual MMLU is 'm_mmlu' group in okapi folder",
+        "available_groups": ["m_mmlu"],
+        "available_subtasks": ["m_mmlu_ar", "m_mmlu_de", "m_mmlu_es", "m_mmlu_fr", "etc."]
+      },
+      {
+        "our_name": "okapi",
+        "folder": "lm_eval/tasks/okapi",
+        "note": "Folder exists but no 'okapi' group. Contains multilingual variants of benchmarks",
+        "available_groups": ["m_arc", "m_hellaswag", "m_mmlu", "m_truthfulqa"],
+        "subfolders": ["arc_multilingual", "hellaswag_multilingual", "mmlu_multilingual", "truthfulqa_multilingual"]
+      },
+      {
+        "our_name": "sglue",
+        "folder": "lm_eval/tasks/super_glue",
+        "note": "No 'sglue' group. Super GLUE tasks are registered individually or as 'super-glue-lm-eval-v1'",
+        "available_groups": ["super-glue-lm-eval-v1", "super-glue-lm-eval-v1-seq2seq", "super-glue-t5-prompt"],
+        "available_subtasks": ["boolq", "cb", "copa", "multirc", "record", "sglue_rte", "wic", "wsc"]
+      },
+      {
+        "our_name": "superglue",
+        "folder": "lm_eval/tasks/super_glue",
+        "note": "Same as sglue - no 'superglue' group",
+        "available_groups": ["super-glue-lm-eval-v1", "super-glue-lm-eval-v1-seq2seq", "super-glue-t5-prompt"]
+      }
+    ]
+  },
+  "truly_not_in_lmeval": {
+    "description": "Task truly does not exist in lm-eval harness - need to move to HF_EXTRACTORS or remove",
+    "repair": "Move to HF_EXTRACTORS with custom data loader, or remove from manifest",
+    "tasks": [
+      "bhs",
+      "bhtc",
+      "cabbq",
+      "chain",
+      "click",
+      "code2text",
+      "coedit",
+      "discrim_eval",
+      "doc",
+      "egyhellaswag",
+      "egymmlu",
+      "epec",
+      "esbbq",
+      "evalita-sp",
+      "flan",
+      "icelandic_winogrande",
+      "libra",
+      "librusec_history",
+      "lm_syneval",
+      "long_context_multiq",
+      "longbenchv2",
+      "matreshka_names",
+      "multiblimp",
+      "multilingual",
+      "passkey",
+      "quac",
+      "random",
+      "reversed",
+      "ru_2wikimultihopqa",
+      "t0",
+      "tinybenchmarks",
+      "truthfulqa-multi",
+      "turblimp_core",
+      "twenty_newsgroups",
+      "vaxx",
+      "wmt2014"
+    ]
+  },
+  "needs_unitxt": {
+    "description": "Requires unitxt package installation",
+    "repair": "pip install unitxt",
+    "tasks": [
+      "20_newsgroups",
+      "ag_news",
+      "argument_topic",
+      "banking77",
+      "claim_stance_topic",
+      "cnn_dailymail",
+      "dbpedia_14",
+      "ethos_binary",
+      "financial_tweets",
+      "law_stack_exchange",
+      "ledgar",
+      "medical_abstracts",
+      "unfair_tos",
+      "unitxt"
+    ]
+  },
+  "import_error": {
+    "description": "Extractor module cannot be imported - likely missing dependencies or syntax error",
+    "repair": "Fix the extractor module import issues",
+    "tasks": [
+      {"name": "african_flores", "module": "flores"},
+      {"name": "afrobench_adr", "module": "afrobench"},
+      {"name": "agieval", "module": "agieval"},
+      {"name": "evalita-mp", "module": "evalita_mp"},
+      {"name": "flores_ca-eu", "module": "flores"},
+      {"name": "super-glue-t5-prompt", "module": "super_glue_t5_prompt"},
+      {"name": "super_glue-wsc-t5-prompt", "module": "super_glue_t5_prompt"},
+      {"name": "trasnlation_all_flores", "module": "flores"}
+    ]
+  },
+  "no_pairs_returned": {
+    "description": "Extractor ran but returned no contrastive pairs - likely data loading or extraction logic issue",
+    "repair": "Debug extractor to ensure pairs are generated from the dataset",
+    "tasks": [
+      "anagrams1",
+      "anagrams2",
+      "assin_entailment",
+      "cabreu_abstractive",
+      "cocoteros_va",
+      "copa_ca",
+      "cycle_letters",
+      "escola",
+      "gpt3_translation_benchmarks",
+      "kmmlu_accounting",
+      "mgsm_direct_eu",
+      "ncb",
+      "norbelebele_p0",
+      "norec_document_p0",
+      "noropenbookqa_nno_p0",
+      "norrewrite_instruct",
+      "norsummarize_instruct",
+      "openbookqa_ca",
+      "parafraseja",
+      "parafrases_gl",
+      "piqa_eu",
+      "qnlieu",
+      "summarization_gl",
+      "tatoeba_eng_nno_p0",
+      "teca",
+      "truthfulqa_gl_mc1",
+      "turkishmmlu_biology",
+      "unscramble",
+      "wikitext103",
+      "wmt16-en-ro",
+      "wmt16-ro-en",
+      "xlsum_es",
+      "xnli_gl"
+    ]
+  },
+  "timeout": {
+    "description": "Task loading exceeded 30 second timeout",
+    "repair": "Increase timeout or optimize data loading",
+    "tasks": [
+      "bigbench_generate_until",
+      "portuguese_bench",
+      "scrolls"
+    ]
+  },
+  "missing_dependencies": {
+    "description": "Missing Python packages required by lm-eval tasks",
+    "repair_commands": {
+      "tinyBenchmarks": "pip install git+https://github.com/felipemaiapolo/tinyBenchmarks",
+      "langdetect": "pip install langdetect",
+      "emoji": "pip install lm_eval[japanese_leaderboard]",
+      "jsonschema": "pip install jsonschema[format]",
+      "longbench": "pip install lm_eval[longbench]",
+      "bert_score": "pip install evaluate bert-score",
+      "minerva_math": "pip install sympy math_verify antlr4-python3-runtime==4.11",
+      "ruler": "pip install lm_eval[ruler]",
+      "noreval": "pip install sacrebleu bert_score rouge_score"
+    },
+    "tasks": [
+      {"name": "benchmarks", "dep": "tinyBenchmarks"},
+      {"name": "ifeval", "dep": "langdetect"},
+      {"name": "japanese_leaderboard", "dep": "emoji"},
+      {"name": "jsonschema_bench", "dep": "jsonschema"},
+      {"name": "longbench", "dep": "longbench"},
+      {"name": "mediqa_qa2019", "dep": "bert_score"},
+      {"name": "medtext", "dep": "bert_score"},
+      {"name": "meqsum", "dep": "bert_score"},
+      {"name": "mimic_repsum", "dep": "bert_score"},
+      {"name": "minerva_math", "dep": "minerva_math"},
+      {"name": "mts_dialog", "dep": "bert_score"},
+      {"name": "niah_single_1", "dep": "ruler"},
+      {"name": "olaph", "dep": "bert_score"},
+      {"name": "ruler", "dep": "ruler"},
+      {"name": "tinyArc", "dep": "tinyBenchmarks"},
+      {"name": "tinyBenchmarks", "dep": "tinyBenchmarks"},
+      {"name": "tinyGSM8k", "dep": "tinyBenchmarks"},
+      {"name": "tinyHellaswag", "dep": "tinyBenchmarks"},
+      {"name": "tinyMMLU", "dep": "tinyBenchmarks"},
+      {"name": "tinyTruthfulQA", "dep": "tinyBenchmarks"},
+      {"name": "tinyWinogrande", "dep": "tinyBenchmarks"},
+      {"name": "tinyarc", "dep": "tinyBenchmarks"},
+      {"name": "tinygsm8k", "dep": "tinyBenchmarks"},
+      {"name": "tinyhellaswag", "dep": "tinyBenchmarks"},
+      {"name": "tinymmlu", "dep": "tinyBenchmarks"},
+      {"name": "tinytruthfulqa", "dep": "tinyBenchmarks"},
+      {"name": "tinywinogrande", "dep": "tinyBenchmarks"}
+    ]
+  },
+  "other_errors": {
+    "description": "Various other errors requiring individual investigation",
+    "tasks": [
+      {"name": "aradice", "error": "HF dataset file not found"},
+      {"name": "basque_bench", "error": "HF dataset file not found"},
+      {"name": "hle", "error": "Fallback dataset loading not permitted"},
+      {"name": "mbpp", "error": "Multiprocessing bootstrap error"},
+      {"name": "noreval", "error": "Missing sacrebleu, bert_score, rouge_score"},
+      {"name": "norsumm_nno_p0", "error": "Missing sacrebleu, bert_score, rouge_score"},
+      {"name": "pile", "error": "Dataset disabled - the-eye.eu unavailable"},
+      {"name": "social_iqa", "error": "Dataset scripts no longer supported"},
+      {"name": "supergpqa", "error": "Fallback dataset loading not permitted"},
+      {"name": "tmlu", "error": "Extractor interface mismatch - needs lm_eval_task_data argument"}
+    ]
+  },
+  "working_benchmarks": {
+    "count": 179,
+    "examples": [
+      "ArabCulture", "aclue", "acp_bench", "advanced_ai_risk", "afrixnli",
+      "ai2_arc", "anli", "apps", "arabic_leaderboard_acva", "arabicmmlu",
+      "arithmetic", "asdiv", "babi", "bbh", "bbq", "belebele", "blimp",
+      "boolq", "cb", "ceval", "cmmlu_agronomy", "cola", "commonsense_qa",
+      "coqa", "drop", "glue", "gpqa", "gsm8k", "hellaswag", "hendrycks_ethics",
+      "humaneval", "lambada_openai", "logiqa", "math", "mathqa", "medmcqa",
+      "mmlu", "mrpc", "multirc", "mutual", "openbookqa", "piqa", "pubmedqa",
+      "qasper", "race", "sciq", "siqa", "squad_completion", "storycloze",
+      "swag", "triviaqa", "truthfulqa", "winogrande", "wsc273", "xcopa", "xnli"
+    ]
+  }
+}

wisent/parameters/lm_eval/weak_contrastive_pairs.json

@@ -0,0 +1,38 @@
+{
+  "description": "Benchmarks with weak or problematic contrastive pair generation",
+  "categories": {
+    "identical_pairs": {
+      "description": "Positive and negative responses are identical - extractor bug",
+      "benchmarks": [
+        "paloma"
+      ]
+    },
+    "lazy_math_negative": {
+      "description": "Negative is just 'correct_answer + 1' instead of meaningful wrong answer",
+      "benchmarks": [
+        "hendrycks_math",
+        "math500",
+        "livemathbench_cnmo_en",
+        "polymath_en_medium",
+        "polymath_zh_medium",
+        "polymath_en_high",
+        "polymath_zh_high"
+      ]
+    },
+    "hedging_negative": {
+      "description": "Negative is 'I believe the answer is not X' instead of actual wrong answer",
+      "benchmarks": [
+        "simpleqa",
+        "frames"
+      ]
+    },
+    "negation_pattern": {
+      "description": "Negative is 'not X' pattern - acceptable for some benchmarks but weak",
+      "benchmarks": [
+        "babi"
+      ]
+    }
+  },
+  "total_weak_benchmarks": 12,
+  "notes": "These benchmarks technically work but have suboptimal contrastive pair quality. Consider improving extractors to generate more meaningful negative examples."
+}