tp4-switchless-ring-20260915.json 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430
  1. {
  2. "decode_window_idle": {
  3. "captured_utc": "20260915T194011Z",
  4. "base_url_redacted": "http://<head>:8888",
  5. "waves": [
  6. {
  7. "concurrency": 1,
  8. "wall_seconds": 49.321,
  9. "successes": 1,
  10. "completion_tokens": 3651,
  11. "aggregate_tok_s": 74.03,
  12. "median_window_tps": 67.41,
  13. "mean_window_tps": 67.41,
  14. "median_ttft_s": 0.535,
  15. "finish_reasons": [
  16. "stop"
  17. ],
  18. "idle_before": {
  19. "num_running_reqs": 0,
  20. "num_waiting_reqs": 0,
  21. "num_used_tokens": 0
  22. },
  23. "after": {
  24. "num_running_reqs": 0,
  25. "num_waiting_reqs": 0,
  26. "num_used_tokens": 0
  27. },
  28. "per_request": [
  29. {
  30. "index": 0,
  31. "seconds": 49.318,
  32. "ttft_s": 0.535,
  33. "window_129_641_tps": 67.41,
  34. "completion_tokens": 3651,
  35. "finish_reason": "stop"
  36. }
  37. ]
  38. },
  39. {
  40. "concurrency": 2,
  41. "wall_seconds": 61.714,
  42. "successes": 2,
  43. "completion_tokens": 6880,
  44. "aggregate_tok_s": 111.48,
  45. "median_window_tps": 54.98,
  46. "mean_window_tps": 54.98,
  47. "median_ttft_s": 0.738,
  48. "finish_reasons": [
  49. "stop",
  50. "stop"
  51. ],
  52. "idle_before": {
  53. "num_running_reqs": 0,
  54. "num_waiting_reqs": 0,
  55. "num_used_tokens": 0
  56. },
  57. "after": {
  58. "num_running_reqs": 0,
  59. "num_waiting_reqs": 0,
  60. "num_used_tokens": 0
  61. },
  62. "per_request": [
  63. {
  64. "index": 0,
  65. "seconds": 55.228,
  66. "ttft_s": 0.738,
  67. "window_129_641_tps": 55.21,
  68. "completion_tokens": 3190,
  69. "finish_reason": "stop"
  70. },
  71. {
  72. "index": 1,
  73. "seconds": 61.713,
  74. "ttft_s": 0.737,
  75. "window_129_641_tps": 54.74,
  76. "completion_tokens": 3690,
  77. "finish_reason": "stop"
  78. }
  79. ]
  80. },
  81. {
  82. "concurrency": 4,
  83. "wall_seconds": 81.536,
  84. "successes": 4,
  85. "completion_tokens": 13050,
  86. "aggregate_tok_s": 160.05,
  87. "median_window_tps": 40.84,
  88. "mean_window_tps": 40.86,
  89. "median_ttft_s": 0.978,
  90. "finish_reasons": [
  91. "stop",
  92. "stop",
  93. "stop",
  94. "stop"
  95. ],
  96. "idle_before": {
  97. "num_running_reqs": 0,
  98. "num_waiting_reqs": 0,
  99. "num_used_tokens": 0
  100. },
  101. "after": {
  102. "num_running_reqs": 0,
  103. "num_waiting_reqs": 0,
  104. "num_used_tokens": 0
  105. },
  106. "per_request": [
  107. {
  108. "index": 0,
  109. "seconds": 80.28,
  110. "ttft_s": 0.809,
  111. "window_129_641_tps": 41.35,
  112. "completion_tokens": 3216,
  113. "finish_reason": "stop"
  114. },
  115. {
  116. "index": 1,
  117. "seconds": 76.574,
  118. "ttft_s": 1.291,
  119. "window_129_641_tps": 40.42,
  120. "completion_tokens": 3146,
  121. "finish_reason": "stop"
  122. },
  123. {
  124. "index": 2,
  125. "seconds": 81.534,
  126. "ttft_s": 0.918,
  127. "window_129_641_tps": 40.91,
  128. "completion_tokens": 3397,
  129. "finish_reason": "stop"
  130. },
  131. {
  132. "index": 3,
  133. "seconds": 80.171,
  134. "ttft_s": 1.039,
  135. "window_129_641_tps": 40.77,
  136. "completion_tokens": 3291,
  137. "finish_reason": "stop"
  138. }
  139. ]
  140. },
  141. {
  142. "concurrency": 8,
  143. "wall_seconds": 119.922,
  144. "successes": 8,
  145. "completion_tokens": 26963,
  146. "aggregate_tok_s": 224.84,
  147. "median_window_tps": 29.91,
  148. "mean_window_tps": 29.76,
  149. "median_ttft_s": 1.181,
  150. "finish_reasons": [
  151. "stop",
  152. "stop",
  153. "stop",
  154. "stop",
  155. "stop",
  156. "stop",
  157. "stop",
  158. "stop"
  159. ],
  160. "idle_before": {
  161. "num_running_reqs": 0,
  162. "num_waiting_reqs": 0,
  163. "num_used_tokens": 0
  164. },
  165. "after": {
  166. "num_running_reqs": 0,
  167. "num_waiting_reqs": 0,
  168. "num_used_tokens": 0
  169. },
  170. "per_request": [
  171. {
  172. "index": 0,
  173. "seconds": 113.437,
  174. "ttft_s": 1.181,
  175. "window_129_641_tps": 32.64,
  176. "completion_tokens": 3353,
  177. "finish_reason": "stop"
  178. },
  179. {
  180. "index": 1,
  181. "seconds": 110.102,
  182. "ttft_s": 1.181,
  183. "window_129_641_tps": 29.62,
  184. "completion_tokens": 3258,
  185. "finish_reason": "stop"
  186. },
  187. {
  188. "index": 2,
  189. "seconds": 103.041,
  190. "ttft_s": 1.222,
  191. "window_129_641_tps": 30.19,
  192. "completion_tokens": 3013,
  193. "finish_reason": "stop"
  194. },
  195. {
  196. "index": 3,
  197. "seconds": 107.252,
  198. "ttft_s": 1.181,
  199. "window_129_641_tps": 30.61,
  200. "completion_tokens": 3115,
  201. "finish_reason": "stop"
  202. },
  203. {
  204. "index": 4,
  205. "seconds": 116.409,
  206. "ttft_s": 1.181,
  207. "window_129_641_tps": 29.03,
  208. "completion_tokens": 3470,
  209. "finish_reason": "stop"
  210. },
  211. {
  212. "index": 5,
  213. "seconds": 118.41,
  214. "ttft_s": 1.175,
  215. "window_129_641_tps": 27.55,
  216. "completion_tokens": 3585,
  217. "finish_reason": "stop"
  218. },
  219. {
  220. "index": 6,
  221. "seconds": 119.919,
  222. "ttft_s": 1.175,
  223. "window_129_641_tps": 27.71,
  224. "completion_tokens": 3610,
  225. "finish_reason": "stop"
  226. },
  227. {
  228. "index": 7,
  229. "seconds": 117.113,
  230. "ttft_s": 1.18,
  231. "window_129_641_tps": 30.73,
  232. "completion_tokens": 3559,
  233. "finish_reason": "stop"
  234. }
  235. ]
  236. }
  237. ]
  238. },
  239. "openai_correctness": {
  240. "models": {
  241. "object": "list",
  242. "data": [
  243. {
  244. "id": "deepseek-v4.1-flash",
  245. "object": "model",
  246. "created": 1789500134,
  247. "owned_by": "sglang",
  248. "root": "deepseek-v4.1-flash",
  249. "parent": null,
  250. "max_model_len": 1048576
  251. }
  252. ]
  253. },
  254. "health_http": 200,
  255. "arithmetic": {
  256. "http": 200,
  257. "seconds": 1.694,
  258. "finish_reason": "stop",
  259. "content": "42",
  260. "reasoning": "",
  261. "reasoning_tokens": 0,
  262. "prompt_tokens": 18,
  263. "completion_tokens": 2,
  264. "tool_calls": []
  265. },
  266. "thinking": {
  267. "http": 200,
  268. "seconds": 2.685,
  269. "finish_reason": "stop",
  270. "content": "OK99",
  271. "reasoning": "We need answer exactly OK99 and nothing else. Need comply. Final only.",
  272. "reasoning_tokens": 17,
  273. "prompt_tokens": 39,
  274. "completion_tokens": 20,
  275. "tool_calls": []
  276. },
  277. "tools": {
  278. "http": 200,
  279. "seconds": 2.739,
  280. "finish_reason": "tool_calls",
  281. "content": "I'll check the weather in Vigo for you.",
  282. "reasoning": "",
  283. "reasoning_tokens": 0,
  284. "prompt_tokens": 270,
  285. "completion_tokens": 50,
  286. "tool_calls": [
  287. {
  288. "id": "call_9cf65bc578424fff896e7f7c",
  289. "index": 0,
  290. "type": "function",
  291. "function": {
  292. "name": "get_weather",
  293. "arguments": "{\"city\": \"Vigo\"}"
  294. }
  295. }
  296. ]
  297. }
  298. },
  299. "openai_short_prose_contaminated": [
  300. {
  301. "concurrency": 1,
  302. "max_tokens": 256,
  303. "thinking": false,
  304. "wall_seconds": 16.238,
  305. "successes": 1,
  306. "completion_tokens": 256,
  307. "aggregate_tok_s": 15.77,
  308. "mean_per_request_tok_s": 15.77,
  309. "min_per_request_tok_s": 15.77,
  310. "finish_reasons": [
  311. "length"
  312. ],
  313. "seconds": [
  314. 16.238
  315. ]
  316. },
  317. {
  318. "concurrency": 2,
  319. "max_tokens": 256,
  320. "thinking": false,
  321. "wall_seconds": 15.378,
  322. "successes": 2,
  323. "completion_tokens": 512,
  324. "aggregate_tok_s": 33.29,
  325. "mean_per_request_tok_s": 17.47,
  326. "min_per_request_tok_s": 16.65,
  327. "finish_reasons": [
  328. "length",
  329. "length"
  330. ],
  331. "seconds": [
  332. 15.378,
  333. 13.993
  334. ]
  335. },
  336. {
  337. "concurrency": 4,
  338. "max_tokens": 256,
  339. "thinking": false,
  340. "wall_seconds": 39.182,
  341. "successes": 4,
  342. "completion_tokens": 1024,
  343. "aggregate_tok_s": 26.13,
  344. "mean_per_request_tok_s": 6.87,
  345. "min_per_request_tok_s": 6.53,
  346. "finish_reasons": [
  347. "length",
  348. "length",
  349. "length",
  350. "length"
  351. ],
  352. "seconds": [
  353. 39.181,
  354. 36.689,
  355. 36.691,
  356. 36.694
  357. ]
  358. },
  359. {
  360. "concurrency": 8,
  361. "max_tokens": 256,
  362. "thinking": false,
  363. "wall_seconds": 33.406,
  364. "successes": 8,
  365. "completion_tokens": 2048,
  366. "aggregate_tok_s": 61.31,
  367. "mean_per_request_tok_s": 10.2,
  368. "min_per_request_tok_s": 7.66,
  369. "finish_reasons": [
  370. "length",
  371. "length",
  372. "length",
  373. "length",
  374. "length",
  375. "length",
  376. "length",
  377. "length"
  378. ],
  379. "seconds": [
  380. 25.608,
  381. 21.882,
  382. 25.236,
  383. 25.921,
  384. 25.609,
  385. 21.876,
  386. 33.404,
  387. 24.345
  388. ]
  389. }
  390. ],
  391. "needles": [
  392. {
  393. "target_tokens": 32768,
  394. "concurrency": 1,
  395. "calibrated_prompt_tokens": [
  396. 26152
  397. ],
  398. "wall_seconds": 12.797,
  399. "correct": 1
  400. },
  401. {
  402. "target_tokens": 32768,
  403. "concurrency": 4,
  404. "calibrated_prompt_tokens": [
  405. 26152,
  406. 26150,
  407. 26150,
  408. 27442
  409. ],
  410. "wall_seconds": 36.778,
  411. "correct": 4
  412. },
  413. {
  414. "target_tokens": 131072,
  415. "concurrency": 1,
  416. "calibrated_prompt_tokens": [
  417. 107191
  418. ],
  419. "wall_seconds": 414.697,
  420. "correct": 1
  421. }
  422. ],
  423. "long_code_c8": {
  424. "completion_tokens": 48000,
  425. "wall_seconds": 222.413,
  426. "aggregate_tok_s": 215.81,
  427. "successes": 8,
  428. "peak_sglang_decode_tok_s": 258.63
  429. }
  430. }