emergence_results.json 2.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142
  1. {
  2. "cot_reasoning": {
  3. "timestamp": "2026-03-14T22:32:25.567957",
  4. "ability_name": "cot_reasoning",
  5. "r_eff_curve": [
  6. 3.7347538471221924,
  7. 4.651706218719482,
  8. 3.4198639392852783,
  9. 3.5962984561920166,
  10. 6.821524143218994
  11. ],
  12. "model_order": [
  13. "llama3.2-3b-instruct",
  14. "Mistral-7B-v0.3",
  15. "llama3-8b",
  16. "gemma-2-9b-it",
  17. "Qwen2.5-7B-Instruct"
  18. ],
  19. "model_mmlu": [
  20. 58.0,
  21. 62.5,
  22. 68.4,
  23. 82.0,
  24. 86.3
  25. ],
  26. "max_jump": {
  27. "model_index": 3,
  28. "model_name": "gemma-2-9b-it",
  29. "mmlu_at_jump": 82.0,
  30. "jump_magnitude": 3.2252256870269775
  31. },
  32. "changepoints": [
  33. 5
  34. ],
  35. "jump_significance": 2.5551615281279734
  36. },
  37. "arithmetic": {
  38. "timestamp": "2026-03-14T22:35:34.370517",
  39. "ability_name": "arithmetic",
  40. "r_eff_curve": [
  41. 4.50555419921875,
  42. 5.898599624633789,
  43. 5.256784439086914,
  44. 4.236079216003418,
  45. 13.66690731048584
  46. ],
  47. "model_order": [
  48. "llama3.2-3b-instruct",
  49. "Mistral-7B-v0.3",
  50. "llama3-8b",
  51. "gemma-2-9b-it",
  52. "Qwen2.5-7B-Instruct"
  53. ],
  54. "model_mmlu": [
  55. 58.0,
  56. 62.5,
  57. 68.4,
  58. 82.0,
  59. 86.3
  60. ],
  61. "max_jump": {
  62. "model_index": 3,
  63. "model_name": "gemma-2-9b-it",
  64. "mmlu_at_jump": 82.0,
  65. "jump_magnitude": 9.430828094482422
  66. },
  67. "changepoints": [
  68. 5
  69. ],
  70. "jump_significance": 2.6749716525969935
  71. },
  72. "code_generation": {
  73. "timestamp": "2026-03-14T22:40:01.844139",
  74. "ability_name": "code_generation",
  75. "r_eff_curve": [
  76. 6.77747917175293,
  77. 14.093628883361816,
  78. 9.412063598632812,
  79. 7.725470542907715,
  80. 4.795677661895752
  81. ],
  82. "model_order": [
  83. "llama3.2-3b-instruct",
  84. "Mistral-7B-v0.3",
  85. "llama3-8b",
  86. "gemma-2-9b-it",
  87. "Qwen2.5-7B-Instruct"
  88. ],
  89. "model_mmlu": [
  90. 58.0,
  91. 62.5,
  92. 68.4,
  93. 82.0,
  94. 86.3
  95. ],
  96. "max_jump": {
  97. "model_index": 0,
  98. "model_name": "llama3.2-3b-instruct",
  99. "mmlu_at_jump": 58.0,
  100. "jump_magnitude": 7.316149711608887
  101. },
  102. "changepoints": [
  103. 5
  104. ],
  105. "jump_significance": 2.3277893513824783
  106. },
  107. "instruction_following": {
  108. "timestamp": "2026-03-14T22:41:56.391270",
  109. "ability_name": "instruction_following",
  110. "r_eff_curve": [
  111. 7.395653247833252,
  112. 16.79442596435547,
  113. 15.557455062866211,
  114. 11.049053192138672,
  115. 12.142755508422852
  116. ],
  117. "model_order": [
  118. "llama3.2-3b-instruct",
  119. "Mistral-7B-v0.3",
  120. "llama3-8b",
  121. "gemma-2-9b-it",
  122. "Qwen2.5-7B-Instruct"
  123. ],
  124. "model_mmlu": [
  125. 58.0,
  126. 62.5,
  127. 68.4,
  128. 82.0,
  129. 86.3
  130. ],
  131. "max_jump": {
  132. "model_index": 0,
  133. "model_name": "llama3.2-3b-instruct",
  134. "mmlu_at_jump": 58.0,
  135. "jump_magnitude": 9.398772716522217
  136. },
  137. "changepoints": [
  138. 5
  139. ],
  140. "jump_significance": 2.8074494163513304
  141. }
  142. }