model_selection.json 6.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196
  1. {
  2. "all_metrics": {
  3. "codellama:34b": {
  4. "avg_tok_per_sec": 3.2,
  5. "category": "coding",
  6. "coding_composite": 0.437,
  7. "coding_quality": 0.833,
  8. "general_composite": 0.326,
  9. "general_quality": 0.586,
  10. "latency_ms": 4235.4,
  11. "latency_score": 0.153,
  12. "toks_norm": 0.08
  13. },
  14. "deepseek-coder-v2:16b": {
  15. "avg_tok_per_sec": 23.5,
  16. "category": "coding",
  17. "coding_composite": 0.723,
  18. "coding_quality": 0.833,
  19. "general_composite": 0.746,
  20. "general_quality": 0.885,
  21. "latency_ms": 1568.5,
  22. "latency_score": 0.686,
  23. "toks_norm": 0.586
  24. },
  25. "gemma3:12b-it-q4_K_M": {
  26. "avg_tok_per_sec": 6.1,
  27. "category": "general",
  28. "coding_composite": 0.439,
  29. "coding_quality": 0.873,
  30. "general_composite": 0.481,
  31. "general_quality": 0.966,
  32. "latency_ms": 5941.9,
  33. "latency_score": 0,
  34. "toks_norm": 0.153
  35. },
  36. "llama3.1:8b": {
  37. "avg_tok_per_sec": 11.8,
  38. "category": "general",
  39. "coding_composite": 0.599,
  40. "coding_quality": 0.823,
  41. "general_composite": 0.623,
  42. "general_quality": 0.877,
  43. "latency_ms": 2202.0,
  44. "latency_score": 0.56,
  45. "toks_norm": 0.294
  46. },
  47. "llama3.2:3b": {
  48. "avg_tok_per_sec": 23.0,
  49. "category": "general",
  50. "coding_composite": 0.786,
  51. "coding_quality": 0.89,
  52. "general_composite": 0.814,
  53. "general_quality": 0.954,
  54. "latency_ms": 754.8,
  55. "latency_score": 0.849,
  56. "toks_norm": 0.576
  57. },
  58. "qwen2.5-coder:14B": {
  59. "avg_tok_per_sec": 6.6,
  60. "category": "coding",
  61. "coding_composite": 0.57,
  62. "coding_quality": 0.85,
  63. "general_composite": 0.607,
  64. "general_quality": 0.931,
  65. "latency_ms": 2229.7,
  66. "latency_score": 0.554,
  67. "toks_norm": 0.164
  68. },
  69. "qwen2.5-coder:7b": {
  70. "avg_tok_per_sec": 12.5,
  71. "category": "coding",
  72. "coding_composite": 0.655,
  73. "coding_quality": 0.85,
  74. "general_composite": 0.682,
  75. "general_quality": 0.91,
  76. "latency_ms": 1431.0,
  77. "latency_score": 0.714,
  78. "toks_norm": 0.312
  79. }
  80. },
  81. "coding_ranking": [
  82. {
  83. "composite": 0.723,
  84. "metrics": {
  85. "avg_tok_per_sec": 23.5,
  86. "category": "coding",
  87. "coding_composite": 0.723,
  88. "coding_quality": 0.833,
  89. "general_composite": 0.746,
  90. "general_quality": 0.885,
  91. "latency_ms": 1568.5,
  92. "latency_score": 0.686,
  93. "toks_norm": 0.586
  94. },
  95. "name": "deepseek-coder-v2:16b"
  96. },
  97. {
  98. "composite": 0.655,
  99. "metrics": {
  100. "avg_tok_per_sec": 12.5,
  101. "category": "coding",
  102. "coding_composite": 0.655,
  103. "coding_quality": 0.85,
  104. "general_composite": 0.682,
  105. "general_quality": 0.91,
  106. "latency_ms": 1431.0,
  107. "latency_score": 0.714,
  108. "toks_norm": 0.312
  109. },
  110. "name": "qwen2.5-coder:7b"
  111. },
  112. {
  113. "composite": 0.57,
  114. "metrics": {
  115. "avg_tok_per_sec": 6.6,
  116. "category": "coding",
  117. "coding_composite": 0.57,
  118. "coding_quality": 0.85,
  119. "general_composite": 0.607,
  120. "general_quality": 0.931,
  121. "latency_ms": 2229.7,
  122. "latency_score": 0.554,
  123. "toks_norm": 0.164
  124. },
  125. "name": "qwen2.5-coder:14B"
  126. },
  127. {
  128. "composite": 0.437,
  129. "metrics": {
  130. "avg_tok_per_sec": 3.2,
  131. "category": "coding",
  132. "coding_composite": 0.437,
  133. "coding_quality": 0.833,
  134. "general_composite": 0.326,
  135. "general_quality": 0.586,
  136. "latency_ms": 4235.4,
  137. "latency_score": 0.153,
  138. "toks_norm": 0.08
  139. },
  140. "name": "codellama:34b"
  141. }
  142. ],
  143. "general_ranking": [
  144. {
  145. "composite": 0.814,
  146. "metrics": {
  147. "avg_tok_per_sec": 23.0,
  148. "category": "general",
  149. "coding_composite": 0.786,
  150. "coding_quality": 0.89,
  151. "general_composite": 0.814,
  152. "general_quality": 0.954,
  153. "latency_ms": 754.8,
  154. "latency_score": 0.849,
  155. "toks_norm": 0.576
  156. },
  157. "name": "llama3.2:3b"
  158. },
  159. {
  160. "composite": 0.623,
  161. "metrics": {
  162. "avg_tok_per_sec": 11.8,
  163. "category": "general",
  164. "coding_composite": 0.599,
  165. "coding_quality": 0.823,
  166. "general_composite": 0.623,
  167. "general_quality": 0.877,
  168. "latency_ms": 2202.0,
  169. "latency_score": 0.56,
  170. "toks_norm": 0.294
  171. },
  172. "name": "llama3.1:8b"
  173. },
  174. {
  175. "composite": 0.481,
  176. "metrics": {
  177. "avg_tok_per_sec": 6.1,
  178. "category": "general",
  179. "coding_composite": 0.439,
  180. "coding_quality": 0.873,
  181. "general_composite": 0.481,
  182. "general_quality": 0.966,
  183. "latency_ms": 5941.9,
  184. "latency_score": 0,
  185. "toks_norm": 0.153
  186. },
  187. "name": "gemma3:12b-it-q4_K_M"
  188. }
  189. ],
  190. "slot1_general": "llama3.2:3b",
  191. "slot2_general": "llama3.1:8b",
  192. "slot3_coding": "deepseek-coder-v2:16b",
  193. "slot4_coding": "qwen2.5-coder:7b",
  194. "slot5_general_rotate": "gemma3:12b-it-q4_K_M",
  195. "slot6_coding_rotate": "qwen2.5-coder:14B"
  196. }