wisent 0.7.701__py3-none-any.whl → 0.7.901__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (330) hide show
  1. wisent/__init__.py +1 -1
  2. wisent/core/activations/activation_cache.py +393 -0
  3. wisent/core/activations/activations.py +3 -3
  4. wisent/core/activations/activations_collector.py +9 -5
  5. wisent/core/activations/classifier_inference_strategy.py +12 -11
  6. wisent/core/activations/extraction_strategy.py +256 -84
  7. wisent/core/classifiers/classifiers/core/atoms.py +3 -2
  8. wisent/core/cli/__init__.py +2 -1
  9. wisent/core/cli/agent/apply_steering.py +5 -7
  10. wisent/core/cli/agent/train_classifier.py +19 -7
  11. wisent/core/cli/check_linearity.py +35 -3
  12. wisent/core/cli/cluster_benchmarks.py +4 -6
  13. wisent/core/cli/create_steering_vector.py +6 -4
  14. wisent/core/cli/diagnose_vectors.py +7 -4
  15. wisent/core/cli/estimate_unified_goodness_time.py +6 -4
  16. wisent/core/cli/generate_pairs_from_task.py +9 -56
  17. wisent/core/cli/geometry_search.py +137 -0
  18. wisent/core/cli/get_activations.py +1 -1
  19. wisent/core/cli/method_optimizer.py +4 -3
  20. wisent/core/cli/modify_weights.py +3 -2
  21. wisent/core/cli/optimize_sample_size.py +1 -1
  22. wisent/core/cli/optimize_steering.py +14 -16
  23. wisent/core/cli/optimize_weights.py +2 -1
  24. wisent/core/cli/preview_pairs.py +203 -0
  25. wisent/core/cli/steering_method_trainer.py +3 -3
  26. wisent/core/cli/tasks.py +19 -76
  27. wisent/core/cli/train_unified_goodness.py +3 -3
  28. wisent/core/contrastive_pairs/diagnostics/control_vectors.py +4 -4
  29. wisent/core/contrastive_pairs/diagnostics/linearity.py +7 -0
  30. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/agentic_search.py +37 -347
  31. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/aider_polyglot.py +113 -136
  32. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codeforces.py +2 -12
  33. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/coding_benchmarks.py +124 -504
  34. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/faithbench.py +40 -63
  35. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/flames.py +46 -89
  36. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/flores.py +15 -4
  37. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/frames.py +36 -20
  38. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/hallucinations_leaderboard.py +3 -45
  39. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/livemathbench.py +42 -4
  40. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/longform_writing.py +2 -112
  41. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/math500.py +39 -4
  42. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/medium_priority_benchmarks.py +475 -525
  43. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/mercury.py +65 -42
  44. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/olympiadbench.py +2 -12
  45. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/planbench.py +78 -219
  46. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/polymath.py +37 -4
  47. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/recode.py +84 -69
  48. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/refusalbench.py +168 -160
  49. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/simpleqa.py +44 -25
  50. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/tau_bench.py +3 -103
  51. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/toolbench.py +3 -97
  52. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/toolemu.py +48 -182
  53. wisent/core/contrastive_pairs/lm_eval_pairs/lm_extractor_manifest.py +3 -0
  54. wisent/core/contrastive_pairs/lm_eval_pairs/lm_extractor_registry.py +19 -1
  55. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/aclue.py +1 -3
  56. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/acp_bench.py +1 -3
  57. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/acp_bench_hard.py +1 -3
  58. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/advanced.py +2 -4
  59. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/aexams.py +1 -3
  60. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/afrimmlu.py +1 -3
  61. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/afrixnli.py +2 -2
  62. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arabculture.py +1 -3
  63. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arabic.py +1 -3
  64. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arabic_exams.py +1 -3
  65. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arabic_leaderboard_complete.py +1 -3
  66. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arabic_leaderboard_light.py +1 -3
  67. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arabicmmlu.py +1 -3
  68. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/aradice.py +1 -3
  69. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arc.py +1 -3
  70. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arc_challenge.py +1 -2
  71. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arc_easy.py +1 -2
  72. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/arithmetic.py +2 -2
  73. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/asdiv.py +2 -2
  74. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/babi.py +36 -2
  75. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/basque_bench.py +1 -3
  76. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/bbq.py +1 -3
  77. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/belebele.py +1 -3
  78. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/benchmarks.py +1 -3
  79. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/bertaqa.py +1 -3
  80. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/bhs.py +1 -3
  81. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/bhtc.py +3 -5
  82. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/blimp.py +1 -3
  83. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/blimp_nl.py +1 -3
  84. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/boolq.py +2 -2
  85. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/c4.py +1 -3
  86. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/cabbq.py +1 -3
  87. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/careqa.py +1 -3
  88. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/catalan_bench.py +1 -3
  89. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/catalanqa.py +1 -3
  90. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/catcola.py +1 -3
  91. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/cb.py +2 -2
  92. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/ceval.py +1 -3
  93. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/ceval_valid.py +1 -3
  94. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/chain.py +1 -3
  95. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/chartqa.py +1 -3
  96. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/claim.py +1 -3
  97. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/click.py +1 -3
  98. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/cmmlu.py +1 -3
  99. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/cnn.py +1 -3
  100. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/cocoteros.py +1 -3
  101. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/coedit.py +1 -3
  102. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/commonsense.py +1 -3
  103. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/commonsense_qa.py +1 -3
  104. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/copa.py +2 -2
  105. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/copal_id.py +1 -3
  106. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/coqa.py +3 -4
  107. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/csatqa.py +1 -3
  108. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/cycle.py +1 -3
  109. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/darija_bench.py +1 -3
  110. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/darijahellaswag.py +2 -6
  111. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/darijammlu.py +1 -3
  112. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/dbpedia.py +1 -3
  113. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/discrim_eval.py +1 -3
  114. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/doc.py +1 -3
  115. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/drop.py +2 -2
  116. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/epec.py +1 -3
  117. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eq.py +1 -3
  118. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eq_bench.py +1 -3
  119. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eq_bench_ca.py +1 -3
  120. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eq_bench_es.py +1 -3
  121. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/esbbq.py +1 -3
  122. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/ethics.py +1 -3
  123. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eus.py +1 -3
  124. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eus_exams.py +1 -3
  125. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eus_proficiency.py +1 -3
  126. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eus_reading.py +1 -3
  127. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/eus_trivia.py +1 -3
  128. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/evalita_llm.py +1 -3
  129. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/financial.py +1 -3
  130. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/flan.py +1 -3
  131. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/french_bench.py +1 -3
  132. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/galician_bench.py +1 -3
  133. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/gaokao.py +2 -2
  134. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/glianorex.py +1 -3
  135. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/global_mmlu.py +1 -3
  136. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/global_piqa.py +1 -3
  137. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/gpt3.py +1 -3
  138. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/groundcocoa.py +1 -3
  139. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/haerae.py +1 -3
  140. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/headqa.py +2 -2
  141. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/hellaswag.py +2 -2
  142. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/hendrycks_ethics.py +5 -9
  143. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/hendrycks_math.py +63 -16
  144. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/histoires_morales.py +1 -3
  145. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/hrm8k.py +1 -3
  146. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/humaneval_infilling.py +1 -3
  147. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/icelandic_winogrande.py +1 -3
  148. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/inverse.py +1 -3
  149. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/inverse_scaling.py +1 -3
  150. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/ja.py +1 -3
  151. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/japanese_leaderboard.py +1 -3
  152. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/japanese_leaderboard_mc.py +1 -1
  153. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/kmmlu.py +1 -3
  154. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/kobest.py +1 -3
  155. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/kormedmcqa.py +5 -17
  156. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/lambada_cloze.py +1 -3
  157. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/lambada_multilingual.py +1 -3
  158. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/law.py +1 -3
  159. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/leaderboard.py +1 -3
  160. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/lingoly.py +1 -3
  161. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/llama3.py +1 -3
  162. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/lm_syneval.py +1 -3
  163. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/logiqa.py +2 -2
  164. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/logiqa2.py +2 -2
  165. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/longbench.py +1 -3
  166. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/longbenchv2.py +1 -3
  167. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mastermind.py +2 -4
  168. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mc-taco.py +2 -2
  169. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/med_concepts_qa.py +2 -4
  170. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/meddialog.py +1 -3
  171. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/medical.py +1 -3
  172. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/medmcqa.py +1 -3
  173. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/medqa.py +2 -2
  174. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mela.py +2 -2
  175. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/metabench.py +1 -3
  176. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/minerva_math.py +1 -3
  177. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mmlu.py +1 -3
  178. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mmlusr.py +3 -4
  179. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mrpc.py +2 -2
  180. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/multiblimp.py +2 -5
  181. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/multirc.py +2 -2
  182. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mutual.py +2 -2
  183. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/non.py +1 -3
  184. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/noreval.py +1 -3
  185. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/noreval_exact.py +1 -3
  186. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/noreval_gen_exact.py +1 -3
  187. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/noreval_mc.py +4 -8
  188. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/noreval_mc_log_likelihoods.py +4 -8
  189. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/nq_open.py +2 -2
  190. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/okapi_arc_multilingual.py +1 -3
  191. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/okapi_hellaswag_multilingual.py +1 -3
  192. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/okapi_mmlu_multilingual.py +1 -3
  193. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/okapi_truthfulqa_multilingual.py +2 -5
  194. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/olaph.py +1 -3
  195. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/openbookqa.py +2 -2
  196. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/option.py +1 -3
  197. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/parafraseja.py +1 -3
  198. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/parafrases.py +1 -3
  199. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/paws.py +1 -3
  200. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/paws_x.py +1 -3
  201. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/pawsx.py +2 -2
  202. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/persona.py +1 -3
  203. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/phrases.py +1 -3
  204. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/pile.py +1 -3
  205. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/piqa.py +2 -2
  206. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/portuguese_bench.py +1 -3
  207. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/prompt.py +1 -3
  208. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/prost.py +2 -2
  209. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/pubmedqa.py +2 -2
  210. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/qa4mre.py +2 -2
  211. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/qasper.py +2 -2
  212. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/qasper_bool.py +2 -2
  213. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/qnli.py +2 -2
  214. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/qnlieu.py +1 -3
  215. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/qqp.py +2 -2
  216. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/race.py +2 -2
  217. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/random.py +1 -3
  218. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/record.py +2 -2
  219. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/reversed.py +1 -3
  220. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/rte.py +2 -2
  221. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/ruler.py +1 -3
  222. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/sciq.py +2 -2
  223. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/score.py +1 -3
  224. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/scrolls.py +1 -3
  225. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/scrolls_mc.py +1 -3
  226. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/self.py +1 -3
  227. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/sglue.py +1 -3
  228. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/sglue_rte.py +2 -1
  229. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/siqa.py +4 -7
  230. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/social_iqa.py +2 -2
  231. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/spanish_bench.py +1 -3
  232. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/storycloze.py +2 -6
  233. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/summarization.py +1 -3
  234. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/super.py +1 -3
  235. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/super_glue.py +1 -3
  236. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/swag.py +2 -2
  237. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/swde.py +1 -3
  238. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/sycophancy.py +1 -3
  239. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/t0.py +1 -3
  240. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/teca.py +1 -3
  241. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinyarc.py +1 -3
  242. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinybenchmarks.py +1 -3
  243. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinygsm8k.py +1 -3
  244. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinyhellaswag.py +1 -3
  245. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinymmlu.py +1 -3
  246. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinytruthfulqa.py +1 -3
  247. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tinywinogrande.py +1 -3
  248. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/tmmluplus.py +1 -3
  249. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/triviaqa.py +2 -2
  250. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/truthfulqa.py +1 -3
  251. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/truthfulqa_mc1.py +1 -3
  252. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/truthfulqa_mc2.py +1 -3
  253. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/turblimp_core.py +1 -3
  254. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/turkishmmlu.py +1 -3
  255. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/turkishmmlu_mc.py +0 -2
  256. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/unscramble.py +1 -3
  257. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/vaxx.py +2 -2
  258. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/webqs.py +2 -2
  259. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/wic.py +3 -4
  260. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/winogrande.py +2 -2
  261. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/wmdp.py +1 -3
  262. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/wnli.py +2 -2
  263. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/wsc.py +2 -2
  264. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/wsc273.py +1 -3
  265. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/xcopa.py +1 -3
  266. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/xlsum.py +1 -3
  267. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/xnli.py +2 -2
  268. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/xquad.py +2 -4
  269. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/xstorycloze.py +2 -3
  270. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/xwinograd.py +2 -2
  271. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/zhoblimp.py +1 -3
  272. wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_pairs_generation.py +173 -6
  273. wisent/core/data_loaders/loaders/lm_loader.py +12 -1
  274. wisent/core/geometry_runner.py +995 -0
  275. wisent/core/geometry_search_space.py +237 -0
  276. wisent/core/hyperparameter_optimizer.py +1 -1
  277. wisent/core/main.py +3 -0
  278. wisent/core/models/core/atoms.py +5 -3
  279. wisent/core/models/wisent_model.py +1 -1
  280. wisent/core/optuna/classifier/optuna_classifier_optimizer.py +2 -2
  281. wisent/core/parser_arguments/check_linearity_parser.py +12 -2
  282. wisent/core/parser_arguments/generate_vector_from_synthetic_parser.py +2 -2
  283. wisent/core/parser_arguments/generate_vector_from_task_parser.py +2 -2
  284. wisent/core/parser_arguments/geometry_search_parser.py +61 -0
  285. wisent/core/parser_arguments/main_parser.py +8 -0
  286. wisent/core/parser_arguments/train_unified_goodness_parser.py +2 -2
  287. wisent/core/steering.py +5 -3
  288. wisent/core/steering_methods/methods/hyperplane.py +2 -1
  289. wisent/core/synthetic/generators/nonsense_generator.py +30 -18
  290. wisent/core/trainers/steering_trainer.py +2 -2
  291. wisent/core/utils/device.py +27 -27
  292. wisent/core/utils/layer_combinations.py +70 -0
  293. wisent/examples/__init__.py +1 -0
  294. wisent/examples/scripts/__init__.py +1 -0
  295. wisent/examples/scripts/count_all_benchmarks.py +121 -0
  296. wisent/examples/scripts/discover_directions.py +469 -0
  297. wisent/examples/scripts/extract_benchmark_info.py +71 -0
  298. wisent/examples/scripts/generate_paper_data.py +384 -0
  299. wisent/examples/scripts/intervention_validation.py +626 -0
  300. wisent/examples/scripts/results/test_AraDiCE_ArabicMMLU_lev_evaluation.json +324 -0
  301. wisent/examples/scripts/results/test_AraDiCE_ArabicMMLU_lev_pairs.json +92 -0
  302. wisent/examples/scripts/results/test_aexams_IslamicStudies_evaluation.json +324 -0
  303. wisent/examples/scripts/results/test_aexams_IslamicStudies_pairs.json +92 -0
  304. wisent/examples/scripts/results/test_afrimgsm_pairs.json +92 -0
  305. wisent/examples/scripts/results/test_afrimmlu_evaluation.json +324 -0
  306. wisent/examples/scripts/results/test_afrimmlu_pairs.json +92 -0
  307. wisent/examples/scripts/search_all_short_names.py +31 -0
  308. wisent/examples/scripts/test_all_benchmarks.py +138 -0
  309. wisent/examples/scripts/test_all_benchmarks_new.py +28 -0
  310. wisent/examples/scripts/test_contrastive_pairs_all_supported.py +230 -0
  311. wisent/examples/scripts/test_nonsense_baseline.py +261 -0
  312. wisent/examples/scripts/test_one_benchmark.py +324 -0
  313. wisent/examples/scripts/test_one_coding_benchmark.py +293 -0
  314. wisent/examples/scripts/threshold_analysis.py +434 -0
  315. wisent/examples/scripts/visualization_gallery.py +582 -0
  316. wisent/parameters/lm_eval/broken_in_lm_eval.json +179 -2
  317. wisent/parameters/lm_eval/category_directions.json +137 -0
  318. wisent/parameters/lm_eval/repair_plan.json +282 -0
  319. wisent/parameters/lm_eval/weak_contrastive_pairs.json +38 -0
  320. wisent/parameters/lm_eval/working_benchmarks.json +206 -0
  321. wisent/parameters/lm_eval/working_benchmarks_categorized.json +236 -0
  322. wisent/tests/test_detector_accuracy.py +1 -1
  323. wisent/tests/visualize_geometry.py +1 -1
  324. {wisent-0.7.701.dist-info → wisent-0.7.901.dist-info}/METADATA +1 -1
  325. {wisent-0.7.701.dist-info → wisent-0.7.901.dist-info}/RECORD +329 -295
  326. wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/browsecomp.py +0 -245
  327. {wisent-0.7.701.dist-info → wisent-0.7.901.dist-info}/WHEEL +0 -0
  328. {wisent-0.7.701.dist-info → wisent-0.7.901.dist-info}/entry_points.txt +0 -0
  329. {wisent-0.7.701.dist-info → wisent-0.7.901.dist-info}/licenses/LICENSE +0 -0
  330. {wisent-0.7.701.dist-info → wisent-0.7.901.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,324 @@
1
+ {
2
+ "task_name": "AraDiCE_ArabicMMLU_lev",
3
+ "model_name": "mock",
4
+ "evaluator_name": "log_likelihoods",
5
+ "num_pairs": 15,
6
+ "all_correct": true,
7
+ "pairs": [
8
+ {
9
+ "pair_id": 0,
10
+ "prompt": "\u062d\u0643\u0645 \u0645\u0633\u062d \u0627\u0644\u0631\u0643\u0646 \u0627\u0644\u064a\u0645\u0627\u0646\u064a:...",
11
+ "positive_response": "\u0628",
12
+ "negative_response": "\u062c",
13
+ "positive_evaluation": {
14
+ "ground_truth": "TRUTHFUL",
15
+ "confidence": 1.0,
16
+ "expected": "TRUTHFUL",
17
+ "correct": true,
18
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u0628'"
19
+ },
20
+ "negative_evaluation": {
21
+ "ground_truth": "UNTRUTHFUL",
22
+ "confidence": 0.0,
23
+ "expected": "UNTRUTHFUL",
24
+ "correct": true,
25
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u062c'"
26
+ },
27
+ "both_correct": true
28
+ },
29
+ {
30
+ "pair_id": 1,
31
+ "prompt": "\u0628\u062a\u062d\u0648\u0644 \u0627\u0644\u0625\u0634\u0627\u0631\u0627\u062a \u0627\u0644\u0643\u0647\u0631\u0628\u0627\u0626\u064a\u0629 \u0644\u0645\u0648\u062c\u0627\u062a \u0635\u0648\u062a\u064a\u0629:...",
32
+ "positive_response": "\u062f",
33
+ "negative_response": "\u0623",
34
+ "positive_evaluation": {
35
+ "ground_truth": "TRUTHFUL",
36
+ "confidence": 1.0,
37
+ "expected": "TRUTHFUL",
38
+ "correct": true,
39
+ "details": "Predicted: '\u062f' (log_prob=-0.500), Expected: '\u062f'"
40
+ },
41
+ "negative_evaluation": {
42
+ "ground_truth": "UNTRUTHFUL",
43
+ "confidence": 0.0,
44
+ "expected": "UNTRUTHFUL",
45
+ "correct": true,
46
+ "details": "Predicted: '\u062f' (log_prob=-0.500), Expected: '\u0623'"
47
+ },
48
+ "both_correct": true
49
+ },
50
+ {
51
+ "pair_id": 2,
52
+ "prompt": "\u0648\u062d\u062f\u0629 \u0645\u0646 \u0627\u0644\u0634\u063a\u0644\u0627\u062a \u0627\u0644\u062c\u0627\u064a\u0629 \u062a\u0639\u062f \u0645\u0646 \u0622\u062f\u0627\u0628 \u0627\u0644\u062d\u0643\u064a \u0645\u064a\u0632\u0647...",
53
+ "positive_response": "\u062c",
54
+ "negative_response": "\u062f",
55
+ "positive_evaluation": {
56
+ "ground_truth": "TRUTHFUL",
57
+ "confidence": 1.0,
58
+ "expected": "TRUTHFUL",
59
+ "correct": true,
60
+ "details": "Predicted: '\u062c' (log_prob=-0.500), Expected: '\u062c'"
61
+ },
62
+ "negative_evaluation": {
63
+ "ground_truth": "UNTRUTHFUL",
64
+ "confidence": 0.0,
65
+ "expected": "UNTRUTHFUL",
66
+ "correct": true,
67
+ "details": "Predicted: '\u062c' (log_prob=-0.500), Expected: '\u062f'"
68
+ },
69
+ "both_correct": true
70
+ },
71
+ {
72
+ "pair_id": 3,
73
+ "prompt": "\u0627\u0644\u0642\u0637\u0639\u0629 \u0627\u0644\u0625\u0644\u0643\u062a\u0631\u0648\u0646\u064a\u0629 \u0627\u0644\u0644\u064a \u0627\u0633\u062a\u0639\u0645\u0644\u0648\u0647\u0627 \u0628\u0627\u0644\u0645\u0631\u062d\u0644\u0629 \u0627\u0644\u0623\u0648\u0644\u0649 \u0644\u062a\u0637\u0648\u0631 \u0627\u0644\u0643\u0645\u0628\u064a\u0648\u062a\u0631 \u0647\u064a\u064a \u0627\u0644\u062f\u0627\u0631\u0627\u062a \u0627\u0644\u0645\u062a\u0643\u0627\u0645\u0644\u0629....",
74
+ "positive_response": "\u0628",
75
+ "negative_response": "\u062c",
76
+ "positive_evaluation": {
77
+ "ground_truth": "TRUTHFUL",
78
+ "confidence": 1.0,
79
+ "expected": "TRUTHFUL",
80
+ "correct": true,
81
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u0628'"
82
+ },
83
+ "negative_evaluation": {
84
+ "ground_truth": "UNTRUTHFUL",
85
+ "confidence": 0.0,
86
+ "expected": "UNTRUTHFUL",
87
+ "correct": true,
88
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u062c'"
89
+ },
90
+ "both_correct": true
91
+ },
92
+ {
93
+ "pair_id": 4,
94
+ "prompt": "\"\u062d\u0627\u0641\u0638 \u0639\u0644\u0649 ......... \u0644\u0648\u0637\u0646\u0643\". \u062d\u062f\u062f \u0645\u0646 \u0647\u0648\u0646 \u0627\u0644\u062a\u0643\u0645\u0644\u0629 \u0627\u0644\u0635\u062d \u0644\u0644\u0641\u0631\u0627\u063a \u0628\u0627\u0644\u062c\u0645\u0644\u0629 \u0627\u0644\u0633\u0627\u0628\u0642\u0629:...",
95
+ "positive_response": "\u0623",
96
+ "negative_response": "\u0628",
97
+ "positive_evaluation": {
98
+ "ground_truth": "TRUTHFUL",
99
+ "confidence": 1.0,
100
+ "expected": "TRUTHFUL",
101
+ "correct": true,
102
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0623'"
103
+ },
104
+ "negative_evaluation": {
105
+ "ground_truth": "UNTRUTHFUL",
106
+ "confidence": 0.0,
107
+ "expected": "UNTRUTHFUL",
108
+ "correct": true,
109
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0628'"
110
+ },
111
+ "both_correct": true
112
+ },
113
+ {
114
+ "pair_id": 5,
115
+ "prompt": "\u0627\u0633\u062a\u0647\u0627\u0646 \u062e\u0648\u0631\u0634\u064a\u062f \u0628\u0642\u064a\u0645\u0629:...",
116
+ "positive_response": "\u0628",
117
+ "negative_response": "\u062c",
118
+ "positive_evaluation": {
119
+ "ground_truth": "TRUTHFUL",
120
+ "confidence": 1.0,
121
+ "expected": "TRUTHFUL",
122
+ "correct": true,
123
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u0628'"
124
+ },
125
+ "negative_evaluation": {
126
+ "ground_truth": "UNTRUTHFUL",
127
+ "confidence": 0.0,
128
+ "expected": "UNTRUTHFUL",
129
+ "correct": true,
130
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u062c'"
131
+ },
132
+ "both_correct": true
133
+ },
134
+ {
135
+ "pair_id": 6,
136
+ "prompt": "\u0627\u0644\u064a\u0648\u0645 \u0627\u0644\u0633\u0627\u0628\u0642 \u0644\u0644\u0623\u062d\u062f:...",
137
+ "positive_response": "\u0623",
138
+ "negative_response": "\u0628",
139
+ "positive_evaluation": {
140
+ "ground_truth": "TRUTHFUL",
141
+ "confidence": 1.0,
142
+ "expected": "TRUTHFUL",
143
+ "correct": true,
144
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0623'"
145
+ },
146
+ "negative_evaluation": {
147
+ "ground_truth": "UNTRUTHFUL",
148
+ "confidence": 0.0,
149
+ "expected": "UNTRUTHFUL",
150
+ "correct": true,
151
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0628'"
152
+ },
153
+ "both_correct": true
154
+ },
155
+ {
156
+ "pair_id": 7,
157
+ "prompt": "\u062a\u0635\u062d\u0631 \u0627\u0644\u0623\u0631\u0627\u0636\u064a \u0627\u0644\u0647\u0627\u0645\u0634\u064a\u0629 \u0648\u0623\u0631\u0627\u0636\u064a \u0627\u0644\u0628\u0627\u062f\u064a\u0629 \u0628\u0627\u0644\u0623\u0631\u062f\u0646 \u0628\u0633\u0628\u0628:...",
158
+ "positive_response": "\u0628",
159
+ "negative_response": "\u062c",
160
+ "positive_evaluation": {
161
+ "ground_truth": "TRUTHFUL",
162
+ "confidence": 1.0,
163
+ "expected": "TRUTHFUL",
164
+ "correct": true,
165
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u0628'"
166
+ },
167
+ "negative_evaluation": {
168
+ "ground_truth": "UNTRUTHFUL",
169
+ "confidence": 0.0,
170
+ "expected": "UNTRUTHFUL",
171
+ "correct": true,
172
+ "details": "Predicted: '\u0628' (log_prob=-0.500), Expected: '\u062c'"
173
+ },
174
+ "both_correct": true
175
+ },
176
+ {
177
+ "pair_id": 8,
178
+ "prompt": "\u0645\u0646 \u0627\u0644\u0623\u0645\u062b\u0644\u0629 \u0639\u0627\u0644\u062a\u063a\u064a\u0631\u0627\u062a \u0627\u0644\u0643\u064a\u0645\u064a\u0627\u0626\u064a\u0629 \u0627\u0644\u0636\u0627\u0631\u0629...",
179
+ "positive_response": "\u0623",
180
+ "negative_response": "\u0628",
181
+ "positive_evaluation": {
182
+ "ground_truth": "TRUTHFUL",
183
+ "confidence": 1.0,
184
+ "expected": "TRUTHFUL",
185
+ "correct": true,
186
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0623'"
187
+ },
188
+ "negative_evaluation": {
189
+ "ground_truth": "UNTRUTHFUL",
190
+ "confidence": 0.0,
191
+ "expected": "UNTRUTHFUL",
192
+ "correct": true,
193
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0628'"
194
+ },
195
+ "both_correct": true
196
+ },
197
+ {
198
+ "pair_id": 9,
199
+ "prompt": "\u0627\u0644\u0633\u0624\u0627\u0644 \u0627\u0644\u0623\u0648\u0644: \u0627\u062e\u062a\u0627\u0631 \u0631\u0642\u0645 (\u0623) \u0625\u0630\u0627 \u0643\u0627\u0646\u062a \u0627\u0644\u0639\u0628\u0627\u0631\u0629 \u0635\u062d\u064a\u062d\u0629\u060c \u0648\u0631\u0642\u0645 (\u0628) \u0625\u0630\u0627 \u0643\u0627\u0646\u062a \u0627\u0644\u0639\u0628\u0627\u0631\u0629 \u063a\u0644\u0637 . \u0627\u0644\u062a\u0648\u0633\u0651\u0639 \u0628\u0627\u0633\u062a\u0639\u0645\u0627\u0644 \u0627\u0644\u062d\u0627\u0633\u0628\u0627\u062a \u0627\u0644\u0625\u0644\u0643\u062a\u0631\u0648\u0646\u064a\u0629 \u0628\u0627\u0644\u0645\u0646\u0634\u0622\u062a \u0623\u062f\u0649 \u0644\u0625\u0636\u0627\u0641\u0629 \u0623\u0647\u0645\u064a\u0629 \u062c\u062f\u064a\u062f\u0629 \u0644\u062f\u0648\u0631 \u0627\u0644\u0645\u0631\u0627\u062c\u0639 \u0627\u0644\u062f\u0627\u062e\u0644\u064a \u0628\u0647\u0627\u0644\u0645\u0646\u0634\u0622\u062a:...",
200
+ "positive_response": "\u0623",
201
+ "negative_response": "\u0628",
202
+ "positive_evaluation": {
203
+ "ground_truth": "TRUTHFUL",
204
+ "confidence": 1.0,
205
+ "expected": "TRUTHFUL",
206
+ "correct": true,
207
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0623'"
208
+ },
209
+ "negative_evaluation": {
210
+ "ground_truth": "UNTRUTHFUL",
211
+ "confidence": 0.0,
212
+ "expected": "UNTRUTHFUL",
213
+ "correct": true,
214
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0628'"
215
+ },
216
+ "both_correct": true
217
+ },
218
+ {
219
+ "pair_id": 10,
220
+ "prompt": "\u0628\u0646\u0637\u0628\u0642 \u0645\u0628\u062f\u0623 \u0627\u0644\u0633\u0628\u0628 \u0627\u0644\u0642\u0631\u064a\u0628 \u0639\u0644\u0649:...",
221
+ "positive_response": "\u062f",
222
+ "negative_response": "\u0623",
223
+ "positive_evaluation": {
224
+ "ground_truth": "TRUTHFUL",
225
+ "confidence": 1.0,
226
+ "expected": "TRUTHFUL",
227
+ "correct": true,
228
+ "details": "Predicted: '\u062f' (log_prob=-0.500), Expected: '\u062f'"
229
+ },
230
+ "negative_evaluation": {
231
+ "ground_truth": "UNTRUTHFUL",
232
+ "confidence": 0.0,
233
+ "expected": "UNTRUTHFUL",
234
+ "correct": true,
235
+ "details": "Predicted: '\u062f' (log_prob=-0.500), Expected: '\u0623'"
236
+ },
237
+ "both_correct": true
238
+ },
239
+ {
240
+ "pair_id": 11,
241
+ "prompt": "\u0633: \u0634\u0648 \u0647\u0648\u064a \u0627\u0644\u062d\u064a\u0648\u0627\u0646 \u064a\u0644\u064a \u0628\u064a\u062a\u0644\u0642\u0628 \u0628\u0623\u0628\u0648 \u0627\u0644\u0623\u062e\u0637\u0644\u061f...",
242
+ "positive_response": "\u062c",
243
+ "negative_response": "\u0623",
244
+ "positive_evaluation": {
245
+ "ground_truth": "TRUTHFUL",
246
+ "confidence": 1.0,
247
+ "expected": "TRUTHFUL",
248
+ "correct": true,
249
+ "details": "Predicted: '\u062c' (log_prob=-0.500), Expected: '\u062c'"
250
+ },
251
+ "negative_evaluation": {
252
+ "ground_truth": "UNTRUTHFUL",
253
+ "confidence": 0.0,
254
+ "expected": "UNTRUTHFUL",
255
+ "correct": true,
256
+ "details": "Predicted: '\u062c' (log_prob=-0.500), Expected: '\u0623'"
257
+ },
258
+ "both_correct": true
259
+ },
260
+ {
261
+ "pair_id": 12,
262
+ "prompt": "\u2026\u2026\u2026\u0639\u0645\u0644\u064a\u0629 \u0627\u0644\u062a\u0641\u0643\u064a\u0631 \u0644\u0625\u062e\u062a\u064a\u0627\u0631 \u0623\u062d\u0633\u0646 \u0627\u0644\u0628\u062f\u0627\u0626\u0644 \u0623\u0648 \u0627\u0644\u062d\u0644\u0648\u0644 \u0627\u0644\u0645\u062a\u0648\u0641\u0631\u0629 \u0628\u0645\u0648\u0642\u0641....",
263
+ "positive_response": "\u0623",
264
+ "negative_response": "\u0628",
265
+ "positive_evaluation": {
266
+ "ground_truth": "TRUTHFUL",
267
+ "confidence": 1.0,
268
+ "expected": "TRUTHFUL",
269
+ "correct": true,
270
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0623'"
271
+ },
272
+ "negative_evaluation": {
273
+ "ground_truth": "UNTRUTHFUL",
274
+ "confidence": 0.0,
275
+ "expected": "UNTRUTHFUL",
276
+ "correct": true,
277
+ "details": "Predicted: '\u0623' (log_prob=-0.500), Expected: '\u0628'"
278
+ },
279
+ "both_correct": true
280
+ },
281
+ {
282
+ "pair_id": 13,
283
+ "prompt": "\u0643\u0644 \u0627\u0644\u062f\u0648\u0644 \u0627\u0644\u062a\u0627\u0644\u064a\u0629 \u064a\u062d\u062f\u062b \u0628\u064a\u0646\u0647\u0627 \u062a\u0639\u0627\u0643\u0633 \u0641\u0635\u0648\u0644 \u0645\u0627 \u0639\u062f\u0627 :...",
284
+ "positive_response": "\u062f",
285
+ "negative_response": "\u0623",
286
+ "positive_evaluation": {
287
+ "ground_truth": "TRUTHFUL",
288
+ "confidence": 1.0,
289
+ "expected": "TRUTHFUL",
290
+ "correct": true,
291
+ "details": "Predicted: '\u062f' (log_prob=-0.500), Expected: '\u062f'"
292
+ },
293
+ "negative_evaluation": {
294
+ "ground_truth": "UNTRUTHFUL",
295
+ "confidence": 0.0,
296
+ "expected": "UNTRUTHFUL",
297
+ "correct": true,
298
+ "details": "Predicted: '\u062f' (log_prob=-0.500), Expected: '\u0623'"
299
+ },
300
+ "both_correct": true
301
+ },
302
+ {
303
+ "pair_id": 14,
304
+ "prompt": "\u0627\u0646\u0641\u062c\u0631 \u062c\u0633\u0645 \u0643\u062a\u0644\u062a\u0648 \u0643 \u0648\u0633\u0631\u0639\u062a\u0648 \u0639 \u0644\u062c\u0632\u0626\u064a\u064a\u0646 \u0645\u062a\u0633\u0627\u0648\u064a\u064a\u0646 \u0641\u0633\u0631\u0639\u0629 \u0645\u0631\u0643\u0632 \u0627\u0644\u0643\u062a\u0644\u0629 \u0628\u0639\u062f \u0627\u0644\u0627\u0646\u0641\u062c\u0627\u0631 \u0628\u062a\u0633\u0627\u0648\u064a :...",
305
+ "positive_response": "\u062c",
306
+ "negative_response": "\u062f",
307
+ "positive_evaluation": {
308
+ "ground_truth": "TRUTHFUL",
309
+ "confidence": 1.0,
310
+ "expected": "TRUTHFUL",
311
+ "correct": true,
312
+ "details": "Predicted: '\u062c' (log_prob=-0.500), Expected: '\u062c'"
313
+ },
314
+ "negative_evaluation": {
315
+ "ground_truth": "UNTRUTHFUL",
316
+ "confidence": 0.0,
317
+ "expected": "UNTRUTHFUL",
318
+ "correct": true,
319
+ "details": "Predicted: '\u062c' (log_prob=-0.500), Expected: '\u062f'"
320
+ },
321
+ "both_correct": true
322
+ }
323
+ ]
324
+ }
@@ -0,0 +1,92 @@
1
+ [
2
+ {
3
+ "pair_id": 0,
4
+ "prompt": "\u062d\u0643\u0645 \u0645\u0633\u062d \u0627\u0644\u0631\u0643\u0646 \u0627\u0644\u064a\u0645\u0627\u0646\u064a:",
5
+ "positive_response": "\u0628",
6
+ "negative_response": "\u062c"
7
+ },
8
+ {
9
+ "pair_id": 1,
10
+ "prompt": "\u0628\u062a\u062d\u0648\u0644 \u0627\u0644\u0625\u0634\u0627\u0631\u0627\u062a \u0627\u0644\u0643\u0647\u0631\u0628\u0627\u0626\u064a\u0629 \u0644\u0645\u0648\u062c\u0627\u062a \u0635\u0648\u062a\u064a\u0629:",
11
+ "positive_response": "\u062f",
12
+ "negative_response": "\u0623"
13
+ },
14
+ {
15
+ "pair_id": 2,
16
+ "prompt": "\u0648\u062d\u062f\u0629 \u0645\u0646 \u0627\u0644\u0634\u063a\u0644\u0627\u062a \u0627\u0644\u062c\u0627\u064a\u0629 \u062a\u0639\u062f \u0645\u0646 \u0622\u062f\u0627\u0628 \u0627\u0644\u062d\u0643\u064a \u0645\u064a\u0632\u0647",
17
+ "positive_response": "\u062c",
18
+ "negative_response": "\u062f"
19
+ },
20
+ {
21
+ "pair_id": 3,
22
+ "prompt": "\u0627\u0644\u0642\u0637\u0639\u0629 \u0627\u0644\u0625\u0644\u0643\u062a\u0631\u0648\u0646\u064a\u0629 \u0627\u0644\u0644\u064a \u0627\u0633\u062a\u0639\u0645\u0644\u0648\u0647\u0627 \u0628\u0627\u0644\u0645\u0631\u062d\u0644\u0629 \u0627\u0644\u0623\u0648\u0644\u0649 \u0644\u062a\u0637\u0648\u0631 \u0627\u0644\u0643\u0645\u0628\u064a\u0648\u062a\u0631 \u0647\u064a\u064a \u0627\u0644\u062f\u0627\u0631\u0627\u062a \u0627\u0644\u0645\u062a\u0643\u0627\u0645\u0644\u0629.",
23
+ "positive_response": "\u0628",
24
+ "negative_response": "\u062c"
25
+ },
26
+ {
27
+ "pair_id": 4,
28
+ "prompt": "\"\u062d\u0627\u0641\u0638 \u0639\u0644\u0649 ......... \u0644\u0648\u0637\u0646\u0643\". \u062d\u062f\u062f \u0645\u0646 \u0647\u0648\u0646 \u0627\u0644\u062a\u0643\u0645\u0644\u0629 \u0627\u0644\u0635\u062d \u0644\u0644\u0641\u0631\u0627\u063a \u0628\u0627\u0644\u062c\u0645\u0644\u0629 \u0627\u0644\u0633\u0627\u0628\u0642\u0629:",
29
+ "positive_response": "\u0623",
30
+ "negative_response": "\u0628"
31
+ },
32
+ {
33
+ "pair_id": 5,
34
+ "prompt": "\u0627\u0633\u062a\u0647\u0627\u0646 \u062e\u0648\u0631\u0634\u064a\u062f \u0628\u0642\u064a\u0645\u0629:",
35
+ "positive_response": "\u0628",
36
+ "negative_response": "\u062c"
37
+ },
38
+ {
39
+ "pair_id": 6,
40
+ "prompt": "\u0627\u0644\u064a\u0648\u0645 \u0627\u0644\u0633\u0627\u0628\u0642 \u0644\u0644\u0623\u062d\u062f:",
41
+ "positive_response": "\u0623",
42
+ "negative_response": "\u0628"
43
+ },
44
+ {
45
+ "pair_id": 7,
46
+ "prompt": "\u062a\u0635\u062d\u0631 \u0627\u0644\u0623\u0631\u0627\u0636\u064a \u0627\u0644\u0647\u0627\u0645\u0634\u064a\u0629 \u0648\u0623\u0631\u0627\u0636\u064a \u0627\u0644\u0628\u0627\u062f\u064a\u0629 \u0628\u0627\u0644\u0623\u0631\u062f\u0646 \u0628\u0633\u0628\u0628:",
47
+ "positive_response": "\u0628",
48
+ "negative_response": "\u062c"
49
+ },
50
+ {
51
+ "pair_id": 8,
52
+ "prompt": "\u0645\u0646 \u0627\u0644\u0623\u0645\u062b\u0644\u0629 \u0639\u0627\u0644\u062a\u063a\u064a\u0631\u0627\u062a \u0627\u0644\u0643\u064a\u0645\u064a\u0627\u0626\u064a\u0629 \u0627\u0644\u0636\u0627\u0631\u0629",
53
+ "positive_response": "\u0623",
54
+ "negative_response": "\u0628"
55
+ },
56
+ {
57
+ "pair_id": 9,
58
+ "prompt": "\u0627\u0644\u0633\u0624\u0627\u0644 \u0627\u0644\u0623\u0648\u0644: \u0627\u062e\u062a\u0627\u0631 \u0631\u0642\u0645 (\u0623) \u0625\u0630\u0627 \u0643\u0627\u0646\u062a \u0627\u0644\u0639\u0628\u0627\u0631\u0629 \u0635\u062d\u064a\u062d\u0629\u060c \u0648\u0631\u0642\u0645 (\u0628) \u0625\u0630\u0627 \u0643\u0627\u0646\u062a \u0627\u0644\u0639\u0628\u0627\u0631\u0629 \u063a\u0644\u0637 . \u0627\u0644\u062a\u0648\u0633\u0651\u0639 \u0628\u0627\u0633\u062a\u0639\u0645\u0627\u0644 \u0627\u0644\u062d\u0627\u0633\u0628\u0627\u062a \u0627\u0644\u0625\u0644\u0643\u062a\u0631\u0648\u0646\u064a\u0629 \u0628\u0627\u0644\u0645\u0646\u0634\u0622\u062a \u0623\u062f\u0649 \u0644\u0625\u0636\u0627\u0641\u0629 \u0623\u0647\u0645\u064a\u0629 \u062c\u062f\u064a\u062f\u0629 \u0644\u062f\u0648\u0631 \u0627\u0644\u0645\u0631\u0627\u062c\u0639 \u0627\u0644\u062f\u0627\u062e\u0644\u064a \u0628\u0647\u0627\u0644\u0645\u0646\u0634\u0622\u062a:",
59
+ "positive_response": "\u0623",
60
+ "negative_response": "\u0628"
61
+ },
62
+ {
63
+ "pair_id": 10,
64
+ "prompt": "\u0628\u0646\u0637\u0628\u0642 \u0645\u0628\u062f\u0623 \u0627\u0644\u0633\u0628\u0628 \u0627\u0644\u0642\u0631\u064a\u0628 \u0639\u0644\u0649:",
65
+ "positive_response": "\u062f",
66
+ "negative_response": "\u0623"
67
+ },
68
+ {
69
+ "pair_id": 11,
70
+ "prompt": "\u0633: \u0634\u0648 \u0647\u0648\u064a \u0627\u0644\u062d\u064a\u0648\u0627\u0646 \u064a\u0644\u064a \u0628\u064a\u062a\u0644\u0642\u0628 \u0628\u0623\u0628\u0648 \u0627\u0644\u0623\u062e\u0637\u0644\u061f",
71
+ "positive_response": "\u062c",
72
+ "negative_response": "\u0623"
73
+ },
74
+ {
75
+ "pair_id": 12,
76
+ "prompt": "\u2026\u2026\u2026\u0639\u0645\u0644\u064a\u0629 \u0627\u0644\u062a\u0641\u0643\u064a\u0631 \u0644\u0625\u062e\u062a\u064a\u0627\u0631 \u0623\u062d\u0633\u0646 \u0627\u0644\u0628\u062f\u0627\u0626\u0644 \u0623\u0648 \u0627\u0644\u062d\u0644\u0648\u0644 \u0627\u0644\u0645\u062a\u0648\u0641\u0631\u0629 \u0628\u0645\u0648\u0642\u0641.",
77
+ "positive_response": "\u0623",
78
+ "negative_response": "\u0628"
79
+ },
80
+ {
81
+ "pair_id": 13,
82
+ "prompt": "\u0643\u0644 \u0627\u0644\u062f\u0648\u0644 \u0627\u0644\u062a\u0627\u0644\u064a\u0629 \u064a\u062d\u062f\u062b \u0628\u064a\u0646\u0647\u0627 \u062a\u0639\u0627\u0643\u0633 \u0641\u0635\u0648\u0644 \u0645\u0627 \u0639\u062f\u0627 :",
83
+ "positive_response": "\u062f",
84
+ "negative_response": "\u0623"
85
+ },
86
+ {
87
+ "pair_id": 14,
88
+ "prompt": "\u0627\u0646\u0641\u062c\u0631 \u062c\u0633\u0645 \u0643\u062a\u0644\u062a\u0648 \u0643 \u0648\u0633\u0631\u0639\u062a\u0648 \u0639 \u0644\u062c\u0632\u0626\u064a\u064a\u0646 \u0645\u062a\u0633\u0627\u0648\u064a\u064a\u0646 \u0641\u0633\u0631\u0639\u0629 \u0645\u0631\u0643\u0632 \u0627\u0644\u0643\u062a\u0644\u0629 \u0628\u0639\u062f \u0627\u0644\u0627\u0646\u0641\u062c\u0627\u0631 \u0628\u062a\u0633\u0627\u0648\u064a :",
89
+ "positive_response": "\u062c",
90
+ "negative_response": "\u062f"
91
+ }
92
+ ]