llama3_8b_partial_offload_split_1main_1rpc.json 28 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138
  1. {
  2. "estimate": {
  3. "items": [
  4. {
  5. "offloadLayers": 0,
  6. "fullOffloaded": false,
  7. "ram": {
  8. "handleLayers": 32,
  9. "handleLastLayer": 31,
  10. "handleOutputLayer": true,
  11. "remote": false,
  12. "position": 0,
  13. "uma": 1170336728,
  14. "nonuma": 1327623128
  15. },
  16. "vrams": [
  17. {
  18. "handleLayers": 0,
  19. "handleLastLayer": -1,
  20. "handleOutputLayer": false,
  21. "remote": false,
  22. "position": 0,
  23. "uma": 0,
  24. "nonuma": 955768832
  25. },
  26. {
  27. "handleLayers": 0,
  28. "handleLastLayer": -1,
  29. "handleOutputLayer": false,
  30. "remote": false,
  31. "position": 0,
  32. "uma": 0,
  33. "nonuma": 857735168
  34. }
  35. ]
  36. },
  37. {
  38. "offloadLayers": 1,
  39. "fullOffloaded": false,
  40. "ram": {
  41. "handleLayers": 31,
  42. "handleLastLayer": 30,
  43. "handleOutputLayer": true,
  44. "remote": false,
  45. "position": 0,
  46. "uma": 1212279768,
  47. "nonuma": 1369566168
  48. },
  49. "vrams": [
  50. {
  51. "handleLayers": 1,
  52. "handleLastLayer": 31,
  53. "handleOutputLayer": false,
  54. "remote": true,
  55. "position": 0,
  56. "uma": 156270592,
  57. "nonuma": 1137209344
  58. },
  59. {
  60. "handleLayers": 0,
  61. "handleLastLayer": -1,
  62. "handleOutputLayer": false,
  63. "remote": false,
  64. "position": 0,
  65. "uma": 0,
  66. "nonuma": 882905088
  67. }
  68. ]
  69. },
  70. {
  71. "offloadLayers": 2,
  72. "fullOffloaded": false,
  73. "ram": {
  74. "handleLayers": 30,
  75. "handleLastLayer": 29,
  76. "handleOutputLayer": true,
  77. "remote": false,
  78. "position": 0,
  79. "uma": 1178725336,
  80. "nonuma": 1336011736
  81. },
  82. "vrams": [
  83. {
  84. "handleLayers": 1,
  85. "handleLastLayer": 30,
  86. "handleOutputLayer": false,
  87. "remote": true,
  88. "position": 0,
  89. "uma": 156270592,
  90. "nonuma": 1137209344
  91. },
  92. {
  93. "handleLayers": 1,
  94. "handleLastLayer": 31,
  95. "handleOutputLayer": false,
  96. "remote": false,
  97. "position": 0,
  98. "uma": 33554432,
  99. "nonuma": 1039175680
  100. }
  101. ]
  102. },
  103. {
  104. "offloadLayers": 3,
  105. "fullOffloaded": false,
  106. "ram": {
  107. "handleLayers": 29,
  108. "handleLastLayer": 28,
  109. "handleOutputLayer": true,
  110. "remote": false,
  111. "position": 0,
  112. "uma": 1145170904,
  113. "nonuma": 1302457304
  114. },
  115. "vrams": [
  116. {
  117. "handleLayers": 2,
  118. "handleLastLayer": 30,
  119. "handleOutputLayer": false,
  120. "remote": true,
  121. "position": 0,
  122. "uma": 312541184,
  123. "nonuma": 1293479936
  124. },
  125. {
  126. "handleLayers": 1,
  127. "handleLastLayer": 31,
  128. "handleOutputLayer": false,
  129. "remote": false,
  130. "position": 0,
  131. "uma": 33554432,
  132. "nonuma": 1039175680
  133. }
  134. ]
  135. },
  136. {
  137. "offloadLayers": 4,
  138. "fullOffloaded": false,
  139. "ram": {
  140. "handleLayers": 28,
  141. "handleLastLayer": 27,
  142. "handleOutputLayer": true,
  143. "remote": false,
  144. "position": 0,
  145. "uma": 1111616472,
  146. "nonuma": 1268902872
  147. },
  148. "vrams": [
  149. {
  150. "handleLayers": 2,
  151. "handleLastLayer": 29,
  152. "handleOutputLayer": false,
  153. "remote": true,
  154. "position": 0,
  155. "uma": 312541184,
  156. "nonuma": 1293479936
  157. },
  158. {
  159. "handleLayers": 2,
  160. "handleLastLayer": 31,
  161. "handleOutputLayer": false,
  162. "remote": false,
  163. "position": 0,
  164. "uma": 67108864,
  165. "nonuma": 1195446272
  166. }
  167. ]
  168. },
  169. {
  170. "offloadLayers": 5,
  171. "fullOffloaded": false,
  172. "ram": {
  173. "handleLayers": 27,
  174. "handleLastLayer": 26,
  175. "handleOutputLayer": true,
  176. "remote": false,
  177. "position": 0,
  178. "uma": 1078062040,
  179. "nonuma": 1235348440
  180. },
  181. "vrams": [
  182. {
  183. "handleLayers": 3,
  184. "handleLastLayer": 29,
  185. "handleOutputLayer": false,
  186. "remote": true,
  187. "position": 0,
  188. "uma": 468811776,
  189. "nonuma": 1449750528
  190. },
  191. {
  192. "handleLayers": 2,
  193. "handleLastLayer": 31,
  194. "handleOutputLayer": false,
  195. "remote": false,
  196. "position": 0,
  197. "uma": 67108864,
  198. "nonuma": 1195446272
  199. }
  200. ]
  201. },
  202. {
  203. "offloadLayers": 6,
  204. "fullOffloaded": false,
  205. "ram": {
  206. "handleLayers": 26,
  207. "handleLastLayer": 25,
  208. "handleOutputLayer": true,
  209. "remote": false,
  210. "position": 0,
  211. "uma": 1044507608,
  212. "nonuma": 1201794008
  213. },
  214. "vrams": [
  215. {
  216. "handleLayers": 3,
  217. "handleLastLayer": 28,
  218. "handleOutputLayer": false,
  219. "remote": true,
  220. "position": 0,
  221. "uma": 468811776,
  222. "nonuma": 1449750528
  223. },
  224. {
  225. "handleLayers": 3,
  226. "handleLastLayer": 31,
  227. "handleOutputLayer": false,
  228. "remote": false,
  229. "position": 0,
  230. "uma": 100663296,
  231. "nonuma": 1351716864
  232. }
  233. ]
  234. },
  235. {
  236. "offloadLayers": 7,
  237. "fullOffloaded": false,
  238. "ram": {
  239. "handleLayers": 25,
  240. "handleLastLayer": 24,
  241. "handleOutputLayer": true,
  242. "remote": false,
  243. "position": 0,
  244. "uma": 1010953176,
  245. "nonuma": 1168239576
  246. },
  247. "vrams": [
  248. {
  249. "handleLayers": 4,
  250. "handleLastLayer": 28,
  251. "handleOutputLayer": false,
  252. "remote": true,
  253. "position": 0,
  254. "uma": 625082368,
  255. "nonuma": 1606021120
  256. },
  257. {
  258. "handleLayers": 3,
  259. "handleLastLayer": 31,
  260. "handleOutputLayer": false,
  261. "remote": false,
  262. "position": 0,
  263. "uma": 100663296,
  264. "nonuma": 1351716864
  265. }
  266. ]
  267. },
  268. {
  269. "offloadLayers": 8,
  270. "fullOffloaded": false,
  271. "ram": {
  272. "handleLayers": 24,
  273. "handleLastLayer": 23,
  274. "handleOutputLayer": true,
  275. "remote": false,
  276. "position": 0,
  277. "uma": 977398744,
  278. "nonuma": 1134685144
  279. },
  280. "vrams": [
  281. {
  282. "handleLayers": 4,
  283. "handleLastLayer": 27,
  284. "handleOutputLayer": false,
  285. "remote": true,
  286. "position": 0,
  287. "uma": 625082368,
  288. "nonuma": 1606021120
  289. },
  290. {
  291. "handleLayers": 4,
  292. "handleLastLayer": 31,
  293. "handleOutputLayer": false,
  294. "remote": false,
  295. "position": 0,
  296. "uma": 134217728,
  297. "nonuma": 1507987456
  298. }
  299. ]
  300. },
  301. {
  302. "offloadLayers": 9,
  303. "fullOffloaded": false,
  304. "ram": {
  305. "handleLayers": 23,
  306. "handleLastLayer": 22,
  307. "handleOutputLayer": true,
  308. "remote": false,
  309. "position": 0,
  310. "uma": 943844312,
  311. "nonuma": 1101130712
  312. },
  313. "vrams": [
  314. {
  315. "handleLayers": 5,
  316. "handleLastLayer": 27,
  317. "handleOutputLayer": false,
  318. "remote": true,
  319. "position": 0,
  320. "uma": 781352960,
  321. "nonuma": 1762291712
  322. },
  323. {
  324. "handleLayers": 4,
  325. "handleLastLayer": 31,
  326. "handleOutputLayer": false,
  327. "remote": false,
  328. "position": 0,
  329. "uma": 134217728,
  330. "nonuma": 1507987456
  331. }
  332. ]
  333. },
  334. {
  335. "offloadLayers": 10,
  336. "fullOffloaded": false,
  337. "ram": {
  338. "handleLayers": 22,
  339. "handleLastLayer": 21,
  340. "handleOutputLayer": true,
  341. "remote": false,
  342. "position": 0,
  343. "uma": 910289880,
  344. "nonuma": 1067576280
  345. },
  346. "vrams": [
  347. {
  348. "handleLayers": 5,
  349. "handleLastLayer": 26,
  350. "handleOutputLayer": false,
  351. "remote": true,
  352. "position": 0,
  353. "uma": 781352960,
  354. "nonuma": 1762291712
  355. },
  356. {
  357. "handleLayers": 5,
  358. "handleLastLayer": 31,
  359. "handleOutputLayer": false,
  360. "remote": false,
  361. "position": 0,
  362. "uma": 167772160,
  363. "nonuma": 1664258048
  364. }
  365. ]
  366. },
  367. {
  368. "offloadLayers": 11,
  369. "fullOffloaded": false,
  370. "ram": {
  371. "handleLayers": 21,
  372. "handleLastLayer": 20,
  373. "handleOutputLayer": true,
  374. "remote": false,
  375. "position": 0,
  376. "uma": 876735448,
  377. "nonuma": 1034021848
  378. },
  379. "vrams": [
  380. {
  381. "handleLayers": 6,
  382. "handleLastLayer": 26,
  383. "handleOutputLayer": false,
  384. "remote": true,
  385. "position": 0,
  386. "uma": 937623552,
  387. "nonuma": 1918562304
  388. },
  389. {
  390. "handleLayers": 5,
  391. "handleLastLayer": 31,
  392. "handleOutputLayer": false,
  393. "remote": false,
  394. "position": 0,
  395. "uma": 167772160,
  396. "nonuma": 1664258048
  397. }
  398. ]
  399. },
  400. {
  401. "offloadLayers": 12,
  402. "fullOffloaded": false,
  403. "ram": {
  404. "handleLayers": 20,
  405. "handleLastLayer": 19,
  406. "handleOutputLayer": true,
  407. "remote": false,
  408. "position": 0,
  409. "uma": 843181016,
  410. "nonuma": 1000467416
  411. },
  412. "vrams": [
  413. {
  414. "handleLayers": 6,
  415. "handleLastLayer": 25,
  416. "handleOutputLayer": false,
  417. "remote": true,
  418. "position": 0,
  419. "uma": 937623552,
  420. "nonuma": 1918562304
  421. },
  422. {
  423. "handleLayers": 6,
  424. "handleLastLayer": 31,
  425. "handleOutputLayer": false,
  426. "remote": false,
  427. "position": 0,
  428. "uma": 201326592,
  429. "nonuma": 1820528640
  430. }
  431. ]
  432. },
  433. {
  434. "offloadLayers": 13,
  435. "fullOffloaded": false,
  436. "ram": {
  437. "handleLayers": 19,
  438. "handleLastLayer": 18,
  439. "handleOutputLayer": true,
  440. "remote": false,
  441. "position": 0,
  442. "uma": 809626584,
  443. "nonuma": 966912984
  444. },
  445. "vrams": [
  446. {
  447. "handleLayers": 7,
  448. "handleLastLayer": 25,
  449. "handleOutputLayer": false,
  450. "remote": true,
  451. "position": 0,
  452. "uma": 1093894144,
  453. "nonuma": 2074832896
  454. },
  455. {
  456. "handleLayers": 6,
  457. "handleLastLayer": 31,
  458. "handleOutputLayer": false,
  459. "remote": false,
  460. "position": 0,
  461. "uma": 201326592,
  462. "nonuma": 1820528640
  463. }
  464. ]
  465. },
  466. {
  467. "offloadLayers": 14,
  468. "fullOffloaded": false,
  469. "ram": {
  470. "handleLayers": 18,
  471. "handleLastLayer": 17,
  472. "handleOutputLayer": true,
  473. "remote": false,
  474. "position": 0,
  475. "uma": 776072152,
  476. "nonuma": 933358552
  477. },
  478. "vrams": [
  479. {
  480. "handleLayers": 7,
  481. "handleLastLayer": 24,
  482. "handleOutputLayer": false,
  483. "remote": true,
  484. "position": 0,
  485. "uma": 1093894144,
  486. "nonuma": 2074832896
  487. },
  488. {
  489. "handleLayers": 7,
  490. "handleLastLayer": 31,
  491. "handleOutputLayer": false,
  492. "remote": false,
  493. "position": 0,
  494. "uma": 234881024,
  495. "nonuma": 1976799232
  496. }
  497. ]
  498. },
  499. {
  500. "offloadLayers": 15,
  501. "fullOffloaded": false,
  502. "ram": {
  503. "handleLayers": 17,
  504. "handleLastLayer": 16,
  505. "handleOutputLayer": true,
  506. "remote": false,
  507. "position": 0,
  508. "uma": 742517720,
  509. "nonuma": 899804120
  510. },
  511. "vrams": [
  512. {
  513. "handleLayers": 8,
  514. "handleLastLayer": 24,
  515. "handleOutputLayer": false,
  516. "remote": true,
  517. "position": 0,
  518. "uma": 1250164736,
  519. "nonuma": 2231103488
  520. },
  521. {
  522. "handleLayers": 7,
  523. "handleLastLayer": 31,
  524. "handleOutputLayer": false,
  525. "remote": false,
  526. "position": 0,
  527. "uma": 234881024,
  528. "nonuma": 1976799232
  529. }
  530. ]
  531. },
  532. {
  533. "offloadLayers": 16,
  534. "fullOffloaded": false,
  535. "ram": {
  536. "handleLayers": 16,
  537. "handleLastLayer": 15,
  538. "handleOutputLayer": true,
  539. "remote": false,
  540. "position": 0,
  541. "uma": 708963288,
  542. "nonuma": 866249688
  543. },
  544. "vrams": [
  545. {
  546. "handleLayers": 8,
  547. "handleLastLayer": 23,
  548. "handleOutputLayer": false,
  549. "remote": true,
  550. "position": 0,
  551. "uma": 1250164736,
  552. "nonuma": 2231103488
  553. },
  554. {
  555. "handleLayers": 8,
  556. "handleLastLayer": 31,
  557. "handleOutputLayer": false,
  558. "remote": false,
  559. "position": 0,
  560. "uma": 268435456,
  561. "nonuma": 2133069824
  562. }
  563. ]
  564. },
  565. {
  566. "offloadLayers": 17,
  567. "fullOffloaded": false,
  568. "ram": {
  569. "handleLayers": 15,
  570. "handleLastLayer": 14,
  571. "handleOutputLayer": true,
  572. "remote": false,
  573. "position": 0,
  574. "uma": 675408856,
  575. "nonuma": 832695256
  576. },
  577. "vrams": [
  578. {
  579. "handleLayers": 9,
  580. "handleLastLayer": 23,
  581. "handleOutputLayer": false,
  582. "remote": true,
  583. "position": 0,
  584. "uma": 1406435328,
  585. "nonuma": 2387374080
  586. },
  587. {
  588. "handleLayers": 8,
  589. "handleLastLayer": 31,
  590. "handleOutputLayer": false,
  591. "remote": false,
  592. "position": 0,
  593. "uma": 268435456,
  594. "nonuma": 2133069824
  595. }
  596. ]
  597. },
  598. {
  599. "offloadLayers": 18,
  600. "fullOffloaded": false,
  601. "ram": {
  602. "handleLayers": 14,
  603. "handleLastLayer": 13,
  604. "handleOutputLayer": true,
  605. "remote": false,
  606. "position": 0,
  607. "uma": 641854424,
  608. "nonuma": 799140824
  609. },
  610. "vrams": [
  611. {
  612. "handleLayers": 9,
  613. "handleLastLayer": 22,
  614. "handleOutputLayer": false,
  615. "remote": true,
  616. "position": 0,
  617. "uma": 1406435328,
  618. "nonuma": 2387374080
  619. },
  620. {
  621. "handleLayers": 9,
  622. "handleLastLayer": 31,
  623. "handleOutputLayer": false,
  624. "remote": false,
  625. "position": 0,
  626. "uma": 301989888,
  627. "nonuma": 2289340416
  628. }
  629. ]
  630. },
  631. {
  632. "offloadLayers": 19,
  633. "fullOffloaded": false,
  634. "ram": {
  635. "handleLayers": 13,
  636. "handleLastLayer": 12,
  637. "handleOutputLayer": true,
  638. "remote": false,
  639. "position": 0,
  640. "uma": 608299992,
  641. "nonuma": 765586392
  642. },
  643. "vrams": [
  644. {
  645. "handleLayers": 10,
  646. "handleLastLayer": 22,
  647. "handleOutputLayer": false,
  648. "remote": true,
  649. "position": 0,
  650. "uma": 1562705920,
  651. "nonuma": 2543644672
  652. },
  653. {
  654. "handleLayers": 9,
  655. "handleLastLayer": 31,
  656. "handleOutputLayer": false,
  657. "remote": false,
  658. "position": 0,
  659. "uma": 301989888,
  660. "nonuma": 2289340416
  661. }
  662. ]
  663. },
  664. {
  665. "offloadLayers": 20,
  666. "fullOffloaded": false,
  667. "ram": {
  668. "handleLayers": 12,
  669. "handleLastLayer": 11,
  670. "handleOutputLayer": true,
  671. "remote": false,
  672. "position": 0,
  673. "uma": 574745560,
  674. "nonuma": 732031960
  675. },
  676. "vrams": [
  677. {
  678. "handleLayers": 10,
  679. "handleLastLayer": 21,
  680. "handleOutputLayer": false,
  681. "remote": true,
  682. "position": 0,
  683. "uma": 1562705920,
  684. "nonuma": 2543644672
  685. },
  686. {
  687. "handleLayers": 10,
  688. "handleLastLayer": 31,
  689. "handleOutputLayer": false,
  690. "remote": false,
  691. "position": 0,
  692. "uma": 335544320,
  693. "nonuma": 2445611008
  694. }
  695. ]
  696. },
  697. {
  698. "offloadLayers": 21,
  699. "fullOffloaded": false,
  700. "ram": {
  701. "handleLayers": 11,
  702. "handleLastLayer": 10,
  703. "handleOutputLayer": true,
  704. "remote": false,
  705. "position": 0,
  706. "uma": 541191128,
  707. "nonuma": 698477528
  708. },
  709. "vrams": [
  710. {
  711. "handleLayers": 11,
  712. "handleLastLayer": 21,
  713. "handleOutputLayer": false,
  714. "remote": true,
  715. "position": 0,
  716. "uma": 1718976512,
  717. "nonuma": 2699915264
  718. },
  719. {
  720. "handleLayers": 10,
  721. "handleLastLayer": 31,
  722. "handleOutputLayer": false,
  723. "remote": false,
  724. "position": 0,
  725. "uma": 335544320,
  726. "nonuma": 2445611008
  727. }
  728. ]
  729. },
  730. {
  731. "offloadLayers": 22,
  732. "fullOffloaded": false,
  733. "ram": {
  734. "handleLayers": 10,
  735. "handleLastLayer": 9,
  736. "handleOutputLayer": true,
  737. "remote": false,
  738. "position": 0,
  739. "uma": 507636696,
  740. "nonuma": 664923096
  741. },
  742. "vrams": [
  743. {
  744. "handleLayers": 11,
  745. "handleLastLayer": 20,
  746. "handleOutputLayer": false,
  747. "remote": true,
  748. "position": 0,
  749. "uma": 1718976512,
  750. "nonuma": 2699915264
  751. },
  752. {
  753. "handleLayers": 11,
  754. "handleLastLayer": 31,
  755. "handleOutputLayer": false,
  756. "remote": false,
  757. "position": 0,
  758. "uma": 369098752,
  759. "nonuma": 2601881600
  760. }
  761. ]
  762. },
  763. {
  764. "offloadLayers": 23,
  765. "fullOffloaded": false,
  766. "ram": {
  767. "handleLayers": 9,
  768. "handleLastLayer": 8,
  769. "handleOutputLayer": true,
  770. "remote": false,
  771. "position": 0,
  772. "uma": 474082264,
  773. "nonuma": 631368664
  774. },
  775. "vrams": [
  776. {
  777. "handleLayers": 12,
  778. "handleLastLayer": 20,
  779. "handleOutputLayer": false,
  780. "remote": true,
  781. "position": 0,
  782. "uma": 1875247104,
  783. "nonuma": 2856185856
  784. },
  785. {
  786. "handleLayers": 11,
  787. "handleLastLayer": 31,
  788. "handleOutputLayer": false,
  789. "remote": false,
  790. "position": 0,
  791. "uma": 369098752,
  792. "nonuma": 2601881600
  793. }
  794. ]
  795. },
  796. {
  797. "offloadLayers": 24,
  798. "fullOffloaded": false,
  799. "ram": {
  800. "handleLayers": 8,
  801. "handleLastLayer": 7,
  802. "handleOutputLayer": true,
  803. "remote": false,
  804. "position": 0,
  805. "uma": 440527832,
  806. "nonuma": 597814232
  807. },
  808. "vrams": [
  809. {
  810. "handleLayers": 12,
  811. "handleLastLayer": 19,
  812. "handleOutputLayer": false,
  813. "remote": true,
  814. "position": 0,
  815. "uma": 1875247104,
  816. "nonuma": 2856185856
  817. },
  818. {
  819. "handleLayers": 12,
  820. "handleLastLayer": 31,
  821. "handleOutputLayer": false,
  822. "remote": false,
  823. "position": 0,
  824. "uma": 402653184,
  825. "nonuma": 2758152192
  826. }
  827. ]
  828. },
  829. {
  830. "offloadLayers": 25,
  831. "fullOffloaded": false,
  832. "ram": {
  833. "handleLayers": 7,
  834. "handleLastLayer": 6,
  835. "handleOutputLayer": true,
  836. "remote": false,
  837. "position": 0,
  838. "uma": 406973400,
  839. "nonuma": 564259800
  840. },
  841. "vrams": [
  842. {
  843. "handleLayers": 13,
  844. "handleLastLayer": 19,
  845. "handleOutputLayer": false,
  846. "remote": true,
  847. "position": 0,
  848. "uma": 2031517696,
  849. "nonuma": 3012456448
  850. },
  851. {
  852. "handleLayers": 12,
  853. "handleLastLayer": 31,
  854. "handleOutputLayer": false,
  855. "remote": false,
  856. "position": 0,
  857. "uma": 402653184,
  858. "nonuma": 2758152192
  859. }
  860. ]
  861. },
  862. {
  863. "offloadLayers": 26,
  864. "fullOffloaded": false,
  865. "ram": {
  866. "handleLayers": 6,
  867. "handleLastLayer": 5,
  868. "handleOutputLayer": true,
  869. "remote": false,
  870. "position": 0,
  871. "uma": 373418968,
  872. "nonuma": 530705368
  873. },
  874. "vrams": [
  875. {
  876. "handleLayers": 13,
  877. "handleLastLayer": 18,
  878. "handleOutputLayer": false,
  879. "remote": true,
  880. "position": 0,
  881. "uma": 2031517696,
  882. "nonuma": 3012456448
  883. },
  884. {
  885. "handleLayers": 13,
  886. "handleLastLayer": 31,
  887. "handleOutputLayer": false,
  888. "remote": false,
  889. "position": 0,
  890. "uma": 436207616,
  891. "nonuma": 2914422784
  892. }
  893. ]
  894. },
  895. {
  896. "offloadLayers": 27,
  897. "fullOffloaded": false,
  898. "ram": {
  899. "handleLayers": 5,
  900. "handleLastLayer": 4,
  901. "handleOutputLayer": true,
  902. "remote": false,
  903. "position": 0,
  904. "uma": 339864536,
  905. "nonuma": 497150936
  906. },
  907. "vrams": [
  908. {
  909. "handleLayers": 14,
  910. "handleLastLayer": 18,
  911. "handleOutputLayer": false,
  912. "remote": true,
  913. "position": 0,
  914. "uma": 2187788288,
  915. "nonuma": 3168727040
  916. },
  917. {
  918. "handleLayers": 13,
  919. "handleLastLayer": 31,
  920. "handleOutputLayer": false,
  921. "remote": false,
  922. "position": 0,
  923. "uma": 436207616,
  924. "nonuma": 2914422784
  925. }
  926. ]
  927. },
  928. {
  929. "offloadLayers": 28,
  930. "fullOffloaded": false,
  931. "ram": {
  932. "handleLayers": 4,
  933. "handleLastLayer": 3,
  934. "handleOutputLayer": true,
  935. "remote": false,
  936. "position": 0,
  937. "uma": 306310104,
  938. "nonuma": 463596504
  939. },
  940. "vrams": [
  941. {
  942. "handleLayers": 14,
  943. "handleLastLayer": 17,
  944. "handleOutputLayer": false,
  945. "remote": true,
  946. "position": 0,
  947. "uma": 2187788288,
  948. "nonuma": 3168727040
  949. },
  950. {
  951. "handleLayers": 14,
  952. "handleLastLayer": 31,
  953. "handleOutputLayer": false,
  954. "remote": false,
  955. "position": 0,
  956. "uma": 469762048,
  957. "nonuma": 3070693376
  958. }
  959. ]
  960. },
  961. {
  962. "offloadLayers": 29,
  963. "fullOffloaded": false,
  964. "ram": {
  965. "handleLayers": 3,
  966. "handleLastLayer": 2,
  967. "handleOutputLayer": true,
  968. "remote": false,
  969. "position": 0,
  970. "uma": 272755672,
  971. "nonuma": 430042072
  972. },
  973. "vrams": [
  974. {
  975. "handleLayers": 15,
  976. "handleLastLayer": 17,
  977. "handleOutputLayer": false,
  978. "remote": true,
  979. "position": 0,
  980. "uma": 2344058880,
  981. "nonuma": 3324997632
  982. },
  983. {
  984. "handleLayers": 14,
  985. "handleLastLayer": 31,
  986. "handleOutputLayer": false,
  987. "remote": false,
  988. "position": 0,
  989. "uma": 469762048,
  990. "nonuma": 3070693376
  991. }
  992. ]
  993. },
  994. {
  995. "offloadLayers": 30,
  996. "fullOffloaded": false,
  997. "ram": {
  998. "handleLayers": 2,
  999. "handleLastLayer": 1,
  1000. "handleOutputLayer": true,
  1001. "remote": false,
  1002. "position": 0,
  1003. "uma": 239201240,
  1004. "nonuma": 396487640
  1005. },
  1006. "vrams": [
  1007. {
  1008. "handleLayers": 15,
  1009. "handleLastLayer": 16,
  1010. "handleOutputLayer": false,
  1011. "remote": true,
  1012. "position": 0,
  1013. "uma": 2344058880,
  1014. "nonuma": 3324997632
  1015. },
  1016. {
  1017. "handleLayers": 15,
  1018. "handleLastLayer": 31,
  1019. "handleOutputLayer": false,
  1020. "remote": false,
  1021. "position": 0,
  1022. "uma": 503316480,
  1023. "nonuma": 3226963968
  1024. }
  1025. ]
  1026. },
  1027. {
  1028. "offloadLayers": 31,
  1029. "fullOffloaded": false,
  1030. "ram": {
  1031. "handleLayers": 1,
  1032. "handleLastLayer": 0,
  1033. "handleOutputLayer": true,
  1034. "remote": false,
  1035. "position": 0,
  1036. "uma": 205646808,
  1037. "nonuma": 362933208
  1038. },
  1039. "vrams": [
  1040. {
  1041. "handleLayers": 16,
  1042. "handleLastLayer": 16,
  1043. "handleOutputLayer": false,
  1044. "remote": true,
  1045. "position": 0,
  1046. "uma": 2500329472,
  1047. "nonuma": 3481268224
  1048. },
  1049. {
  1050. "handleLayers": 15,
  1051. "handleLastLayer": 31,
  1052. "handleOutputLayer": false,
  1053. "remote": false,
  1054. "position": 0,
  1055. "uma": 503316480,
  1056. "nonuma": 3226963968
  1057. }
  1058. ]
  1059. },
  1060. {
  1061. "offloadLayers": 32,
  1062. "fullOffloaded": false,
  1063. "ram": {
  1064. "handleLayers": 0,
  1065. "handleLastLayer": -1,
  1066. "handleOutputLayer": true,
  1067. "remote": false,
  1068. "position": 0,
  1069. "uma": 172092376,
  1070. "nonuma": 329378776
  1071. },
  1072. "vrams": [
  1073. {
  1074. "handleLayers": 16,
  1075. "handleLastLayer": 15,
  1076. "handleOutputLayer": false,
  1077. "remote": true,
  1078. "position": 0,
  1079. "uma": 2500329472,
  1080. "nonuma": 3481268224
  1081. },
  1082. {
  1083. "handleLayers": 16,
  1084. "handleLastLayer": 31,
  1085. "handleOutputLayer": false,
  1086. "remote": false,
  1087. "position": 0,
  1088. "uma": 536870912,
  1089. "nonuma": 3383234560
  1090. }
  1091. ]
  1092. },
  1093. {
  1094. "offloadLayers": 33,
  1095. "fullOffloaded": true,
  1096. "ram": {
  1097. "handleLayers": 0,
  1098. "handleLastLayer": -1,
  1099. "handleOutputLayer": false,
  1100. "remote": false,
  1101. "position": 0,
  1102. "uma": 171562968,
  1103. "nonuma": 328849368
  1104. },
  1105. "vrams": [
  1106. {
  1107. "handleLayers": 17,
  1108. "handleLastLayer": 16,
  1109. "handleOutputLayer": false,
  1110. "remote": true,
  1111. "position": 0,
  1112. "uma": 2656600064,
  1113. "nonuma": 3539505152
  1114. },
  1115. {
  1116. "handleLayers": 15,
  1117. "handleLastLayer": 31,
  1118. "handleOutputLayer": true,
  1119. "remote": false,
  1120. "position": 0,
  1121. "uma": 503845888,
  1122. "nonuma": 3658449920
  1123. }
  1124. ]
  1125. }
  1126. ],
  1127. "type": "model",
  1128. "architecture": "llama",
  1129. "contextSize": 8192,
  1130. "flashAttention": false,
  1131. "noMMap": false,
  1132. "embeddingOnly": false,
  1133. "reranking": false,
  1134. "distributable": true,
  1135. "logicalBatchSize": 2048,
  1136. "physicalBatchSize": 512
  1137. }
  1138. }