NicoNico6
commited on
Commit
·
8397d83
1
Parent(s):
189d546
update
Browse files- model.safetensors +2 -2
- quant_strategy.json +236 -236
model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:66a8903dd448856cdcd7910b652092f03c58b03cc81b443bd6afce394dfad038
|
3 |
+
size 3437638344
|
quant_strategy.json
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
{
|
2 |
"measurement": {
|
3 |
"model.layers.0": {
|
4 |
-
"accuracy": 0.
|
5 |
-
"total_bits":
|
6 |
"q_proj": {
|
7 |
"group_size": {
|
8 |
"2": 64
|
@@ -17,10 +17,10 @@
|
|
17 |
},
|
18 |
"k_proj": {
|
19 |
"group_size": {
|
20 |
-
"
|
21 |
},
|
22 |
"bits": [
|
23 |
-
|
24 |
],
|
25 |
"bits_prop": [
|
26 |
1
|
@@ -41,10 +41,10 @@
|
|
41 |
},
|
42 |
"o_proj": {
|
43 |
"group_size": {
|
44 |
-
"
|
45 |
},
|
46 |
"bits": [
|
47 |
-
|
48 |
],
|
49 |
"bits_prop": [
|
50 |
1
|
@@ -77,10 +77,10 @@
|
|
77 |
},
|
78 |
"down_proj": {
|
79 |
"group_size": {
|
80 |
-
"
|
81 |
},
|
82 |
"bits": [
|
83 |
-
|
84 |
],
|
85 |
"bits_prop": [
|
86 |
1
|
@@ -177,8 +177,8 @@
|
|
177 |
}
|
178 |
},
|
179 |
"model.layers.2": {
|
180 |
-
"accuracy": 0.
|
181 |
-
"total_bits":
|
182 |
"q_proj": {
|
183 |
"group_size": {
|
184 |
"2": 64
|
@@ -193,10 +193,10 @@
|
|
193 |
},
|
194 |
"k_proj": {
|
195 |
"group_size": {
|
196 |
-
"
|
197 |
},
|
198 |
"bits": [
|
199 |
-
|
200 |
],
|
201 |
"bits_prop": [
|
202 |
1
|
@@ -265,8 +265,8 @@
|
|
265 |
}
|
266 |
},
|
267 |
"model.layers.3": {
|
268 |
-
"accuracy": 0.
|
269 |
-
"total_bits":
|
270 |
"q_proj": {
|
271 |
"group_size": {
|
272 |
"2": 64
|
@@ -281,10 +281,10 @@
|
|
281 |
},
|
282 |
"k_proj": {
|
283 |
"group_size": {
|
284 |
-
"
|
285 |
},
|
286 |
"bits": [
|
287 |
-
|
288 |
],
|
289 |
"bits_prop": [
|
290 |
1
|
@@ -305,10 +305,10 @@
|
|
305 |
},
|
306 |
"o_proj": {
|
307 |
"group_size": {
|
308 |
-
"
|
309 |
},
|
310 |
"bits": [
|
311 |
-
|
312 |
],
|
313 |
"bits_prop": [
|
314 |
1
|
@@ -341,10 +341,10 @@
|
|
341 |
},
|
342 |
"down_proj": {
|
343 |
"group_size": {
|
344 |
-
"
|
345 |
},
|
346 |
"bits": [
|
347 |
-
|
348 |
],
|
349 |
"bits_prop": [
|
350 |
1
|
@@ -353,8 +353,8 @@
|
|
353 |
}
|
354 |
},
|
355 |
"model.layers.4": {
|
356 |
-
"accuracy": 0.
|
357 |
-
"total_bits":
|
358 |
"q_proj": {
|
359 |
"group_size": {
|
360 |
"2": 64
|
@@ -429,10 +429,10 @@
|
|
429 |
},
|
430 |
"down_proj": {
|
431 |
"group_size": {
|
432 |
-
"
|
433 |
},
|
434 |
"bits": [
|
435 |
-
|
436 |
],
|
437 |
"bits_prop": [
|
438 |
1
|
@@ -441,14 +441,14 @@
|
|
441 |
}
|
442 |
},
|
443 |
"model.layers.5": {
|
444 |
-
"accuracy": 0.
|
445 |
-
"total_bits":
|
446 |
"q_proj": {
|
447 |
"group_size": {
|
448 |
-
"
|
449 |
},
|
450 |
"bits": [
|
451 |
-
|
452 |
],
|
453 |
"bits_prop": [
|
454 |
1
|
@@ -457,10 +457,10 @@
|
|
457 |
},
|
458 |
"k_proj": {
|
459 |
"group_size": {
|
460 |
-
"
|
461 |
},
|
462 |
"bits": [
|
463 |
-
|
464 |
],
|
465 |
"bits_prop": [
|
466 |
1
|
@@ -481,10 +481,10 @@
|
|
481 |
},
|
482 |
"o_proj": {
|
483 |
"group_size": {
|
484 |
-
"
|
485 |
},
|
486 |
"bits": [
|
487 |
-
|
488 |
],
|
489 |
"bits_prop": [
|
490 |
1
|
@@ -517,10 +517,10 @@
|
|
517 |
},
|
518 |
"down_proj": {
|
519 |
"group_size": {
|
520 |
-
"
|
521 |
},
|
522 |
"bits": [
|
523 |
-
|
524 |
],
|
525 |
"bits_prop": [
|
526 |
1
|
@@ -529,14 +529,14 @@
|
|
529 |
}
|
530 |
},
|
531 |
"model.layers.6": {
|
532 |
-
"accuracy": 0.
|
533 |
-
"total_bits":
|
534 |
"q_proj": {
|
535 |
"group_size": {
|
536 |
-
"
|
537 |
},
|
538 |
"bits": [
|
539 |
-
|
540 |
],
|
541 |
"bits_prop": [
|
542 |
1
|
@@ -545,10 +545,10 @@
|
|
545 |
},
|
546 |
"k_proj": {
|
547 |
"group_size": {
|
548 |
-
"
|
549 |
},
|
550 |
"bits": [
|
551 |
-
|
552 |
],
|
553 |
"bits_prop": [
|
554 |
1
|
@@ -569,10 +569,10 @@
|
|
569 |
},
|
570 |
"o_proj": {
|
571 |
"group_size": {
|
572 |
-
"
|
573 |
},
|
574 |
"bits": [
|
575 |
-
|
576 |
],
|
577 |
"bits_prop": [
|
578 |
1
|
@@ -605,10 +605,10 @@
|
|
605 |
},
|
606 |
"down_proj": {
|
607 |
"group_size": {
|
608 |
-
"
|
609 |
},
|
610 |
"bits": [
|
611 |
-
|
612 |
],
|
613 |
"bits_prop": [
|
614 |
1
|
@@ -617,14 +617,14 @@
|
|
617 |
}
|
618 |
},
|
619 |
"model.layers.7": {
|
620 |
-
"accuracy": 0.
|
621 |
-
"total_bits":
|
622 |
"q_proj": {
|
623 |
"group_size": {
|
624 |
-
"
|
625 |
},
|
626 |
"bits": [
|
627 |
-
|
628 |
],
|
629 |
"bits_prop": [
|
630 |
1
|
@@ -633,10 +633,10 @@
|
|
633 |
},
|
634 |
"k_proj": {
|
635 |
"group_size": {
|
636 |
-
"
|
637 |
},
|
638 |
"bits": [
|
639 |
-
|
640 |
],
|
641 |
"bits_prop": [
|
642 |
1
|
@@ -657,10 +657,10 @@
|
|
657 |
},
|
658 |
"o_proj": {
|
659 |
"group_size": {
|
660 |
-
"
|
661 |
},
|
662 |
"bits": [
|
663 |
-
|
664 |
],
|
665 |
"bits_prop": [
|
666 |
1
|
@@ -693,10 +693,10 @@
|
|
693 |
},
|
694 |
"down_proj": {
|
695 |
"group_size": {
|
696 |
-
"
|
697 |
},
|
698 |
"bits": [
|
699 |
-
|
700 |
],
|
701 |
"bits_prop": [
|
702 |
1
|
@@ -705,8 +705,8 @@
|
|
705 |
}
|
706 |
},
|
707 |
"model.layers.8": {
|
708 |
-
"accuracy": 0.
|
709 |
-
"total_bits":
|
710 |
"q_proj": {
|
711 |
"group_size": {
|
712 |
"2": 64
|
@@ -781,10 +781,10 @@
|
|
781 |
},
|
782 |
"down_proj": {
|
783 |
"group_size": {
|
784 |
-
"
|
785 |
},
|
786 |
"bits": [
|
787 |
-
|
788 |
],
|
789 |
"bits_prop": [
|
790 |
1
|
@@ -793,8 +793,8 @@
|
|
793 |
}
|
794 |
},
|
795 |
"model.layers.9": {
|
796 |
-
"accuracy": 0.
|
797 |
-
"total_bits":
|
798 |
"q_proj": {
|
799 |
"group_size": {
|
800 |
"2": 64
|
@@ -809,10 +809,10 @@
|
|
809 |
},
|
810 |
"k_proj": {
|
811 |
"group_size": {
|
812 |
-
"
|
813 |
},
|
814 |
"bits": [
|
815 |
-
|
816 |
],
|
817 |
"bits_prop": [
|
818 |
1
|
@@ -833,10 +833,10 @@
|
|
833 |
},
|
834 |
"o_proj": {
|
835 |
"group_size": {
|
836 |
-
"
|
837 |
},
|
838 |
"bits": [
|
839 |
-
|
840 |
],
|
841 |
"bits_prop": [
|
842 |
1
|
@@ -869,10 +869,10 @@
|
|
869 |
},
|
870 |
"down_proj": {
|
871 |
"group_size": {
|
872 |
-
"
|
873 |
},
|
874 |
"bits": [
|
875 |
-
|
876 |
],
|
877 |
"bits_prop": [
|
878 |
1
|
@@ -881,8 +881,8 @@
|
|
881 |
}
|
882 |
},
|
883 |
"model.layers.10": {
|
884 |
-
"accuracy": 0.
|
885 |
-
"total_bits":
|
886 |
"q_proj": {
|
887 |
"group_size": {
|
888 |
"2": 64
|
@@ -921,10 +921,10 @@
|
|
921 |
},
|
922 |
"o_proj": {
|
923 |
"group_size": {
|
924 |
-
"
|
925 |
},
|
926 |
"bits": [
|
927 |
-
|
928 |
],
|
929 |
"bits_prop": [
|
930 |
1
|
@@ -957,10 +957,10 @@
|
|
957 |
},
|
958 |
"down_proj": {
|
959 |
"group_size": {
|
960 |
-
"
|
961 |
},
|
962 |
"bits": [
|
963 |
-
|
964 |
],
|
965 |
"bits_prop": [
|
966 |
1
|
@@ -969,8 +969,8 @@
|
|
969 |
}
|
970 |
},
|
971 |
"model.layers.11": {
|
972 |
-
"accuracy": 0.
|
973 |
-
"total_bits":
|
974 |
"q_proj": {
|
975 |
"group_size": {
|
976 |
"2": 64
|
@@ -985,10 +985,10 @@
|
|
985 |
},
|
986 |
"k_proj": {
|
987 |
"group_size": {
|
988 |
-
"
|
989 |
},
|
990 |
"bits": [
|
991 |
-
|
992 |
],
|
993 |
"bits_prop": [
|
994 |
1
|
@@ -1045,10 +1045,10 @@
|
|
1045 |
},
|
1046 |
"down_proj": {
|
1047 |
"group_size": {
|
1048 |
-
"
|
1049 |
},
|
1050 |
"bits": [
|
1051 |
-
|
1052 |
],
|
1053 |
"bits_prop": [
|
1054 |
1
|
@@ -1057,14 +1057,14 @@
|
|
1057 |
}
|
1058 |
},
|
1059 |
"model.layers.12": {
|
1060 |
-
"accuracy": 0.
|
1061 |
-
"total_bits":
|
1062 |
"q_proj": {
|
1063 |
"group_size": {
|
1064 |
-
"
|
1065 |
},
|
1066 |
"bits": [
|
1067 |
-
|
1068 |
],
|
1069 |
"bits_prop": [
|
1070 |
1
|
@@ -1073,10 +1073,10 @@
|
|
1073 |
},
|
1074 |
"k_proj": {
|
1075 |
"group_size": {
|
1076 |
-
"
|
1077 |
},
|
1078 |
"bits": [
|
1079 |
-
|
1080 |
],
|
1081 |
"bits_prop": [
|
1082 |
1
|
@@ -1097,10 +1097,10 @@
|
|
1097 |
},
|
1098 |
"o_proj": {
|
1099 |
"group_size": {
|
1100 |
-
"
|
1101 |
},
|
1102 |
"bits": [
|
1103 |
-
|
1104 |
],
|
1105 |
"bits_prop": [
|
1106 |
1
|
@@ -1133,10 +1133,10 @@
|
|
1133 |
},
|
1134 |
"down_proj": {
|
1135 |
"group_size": {
|
1136 |
-
"
|
1137 |
},
|
1138 |
"bits": [
|
1139 |
-
|
1140 |
],
|
1141 |
"bits_prop": [
|
1142 |
1
|
@@ -1145,8 +1145,8 @@
|
|
1145 |
}
|
1146 |
},
|
1147 |
"model.layers.13": {
|
1148 |
-
"accuracy": 0.
|
1149 |
-
"total_bits":
|
1150 |
"q_proj": {
|
1151 |
"group_size": {
|
1152 |
"2": 64
|
@@ -1161,10 +1161,10 @@
|
|
1161 |
},
|
1162 |
"k_proj": {
|
1163 |
"group_size": {
|
1164 |
-
"
|
1165 |
},
|
1166 |
"bits": [
|
1167 |
-
|
1168 |
],
|
1169 |
"bits_prop": [
|
1170 |
1
|
@@ -1221,10 +1221,10 @@
|
|
1221 |
},
|
1222 |
"down_proj": {
|
1223 |
"group_size": {
|
1224 |
-
"
|
1225 |
},
|
1226 |
"bits": [
|
1227 |
-
|
1228 |
],
|
1229 |
"bits_prop": [
|
1230 |
1
|
@@ -1233,8 +1233,8 @@
|
|
1233 |
}
|
1234 |
},
|
1235 |
"model.layers.14": {
|
1236 |
-
"accuracy": 0.
|
1237 |
-
"total_bits":
|
1238 |
"q_proj": {
|
1239 |
"group_size": {
|
1240 |
"2": 64
|
@@ -1309,10 +1309,10 @@
|
|
1309 |
},
|
1310 |
"down_proj": {
|
1311 |
"group_size": {
|
1312 |
-
"
|
1313 |
},
|
1314 |
"bits": [
|
1315 |
-
|
1316 |
],
|
1317 |
"bits_prop": [
|
1318 |
1
|
@@ -1321,14 +1321,14 @@
|
|
1321 |
}
|
1322 |
},
|
1323 |
"model.layers.15": {
|
1324 |
-
"accuracy": 0.
|
1325 |
-
"total_bits":
|
1326 |
"q_proj": {
|
1327 |
"group_size": {
|
1328 |
-
"
|
1329 |
},
|
1330 |
"bits": [
|
1331 |
-
|
1332 |
],
|
1333 |
"bits_prop": [
|
1334 |
1
|
@@ -1337,10 +1337,10 @@
|
|
1337 |
},
|
1338 |
"k_proj": {
|
1339 |
"group_size": {
|
1340 |
-
"
|
1341 |
},
|
1342 |
"bits": [
|
1343 |
-
|
1344 |
],
|
1345 |
"bits_prop": [
|
1346 |
1
|
@@ -1397,10 +1397,10 @@
|
|
1397 |
},
|
1398 |
"down_proj": {
|
1399 |
"group_size": {
|
1400 |
-
"
|
1401 |
},
|
1402 |
"bits": [
|
1403 |
-
|
1404 |
],
|
1405 |
"bits_prop": [
|
1406 |
1
|
@@ -1409,14 +1409,14 @@
|
|
1409 |
}
|
1410 |
},
|
1411 |
"model.layers.16": {
|
1412 |
-
"accuracy": 0.
|
1413 |
-
"total_bits":
|
1414 |
"q_proj": {
|
1415 |
"group_size": {
|
1416 |
-
"
|
1417 |
},
|
1418 |
"bits": [
|
1419 |
-
|
1420 |
],
|
1421 |
"bits_prop": [
|
1422 |
1
|
@@ -1425,10 +1425,10 @@
|
|
1425 |
},
|
1426 |
"k_proj": {
|
1427 |
"group_size": {
|
1428 |
-
"
|
1429 |
},
|
1430 |
"bits": [
|
1431 |
-
|
1432 |
],
|
1433 |
"bits_prop": [
|
1434 |
1
|
@@ -1485,10 +1485,10 @@
|
|
1485 |
},
|
1486 |
"down_proj": {
|
1487 |
"group_size": {
|
1488 |
-
"
|
1489 |
},
|
1490 |
"bits": [
|
1491 |
-
|
1492 |
],
|
1493 |
"bits_prop": [
|
1494 |
1
|
@@ -1497,8 +1497,8 @@
|
|
1497 |
}
|
1498 |
},
|
1499 |
"model.layers.17": {
|
1500 |
-
"accuracy": 0.
|
1501 |
-
"total_bits":
|
1502 |
"q_proj": {
|
1503 |
"group_size": {
|
1504 |
"2": 64
|
@@ -1513,10 +1513,10 @@
|
|
1513 |
},
|
1514 |
"k_proj": {
|
1515 |
"group_size": {
|
1516 |
-
"
|
1517 |
},
|
1518 |
"bits": [
|
1519 |
-
|
1520 |
],
|
1521 |
"bits_prop": [
|
1522 |
1
|
@@ -1537,10 +1537,10 @@
|
|
1537 |
},
|
1538 |
"o_proj": {
|
1539 |
"group_size": {
|
1540 |
-
"
|
1541 |
},
|
1542 |
"bits": [
|
1543 |
-
|
1544 |
],
|
1545 |
"bits_prop": [
|
1546 |
1
|
@@ -1573,10 +1573,10 @@
|
|
1573 |
},
|
1574 |
"down_proj": {
|
1575 |
"group_size": {
|
1576 |
-
"
|
1577 |
},
|
1578 |
"bits": [
|
1579 |
-
|
1580 |
],
|
1581 |
"bits_prop": [
|
1582 |
1
|
@@ -1585,14 +1585,14 @@
|
|
1585 |
}
|
1586 |
},
|
1587 |
"model.layers.18": {
|
1588 |
-
"accuracy": 0.
|
1589 |
-
"total_bits":
|
1590 |
"q_proj": {
|
1591 |
"group_size": {
|
1592 |
-
"
|
1593 |
},
|
1594 |
"bits": [
|
1595 |
-
|
1596 |
],
|
1597 |
"bits_prop": [
|
1598 |
1
|
@@ -1601,10 +1601,10 @@
|
|
1601 |
},
|
1602 |
"k_proj": {
|
1603 |
"group_size": {
|
1604 |
-
"
|
1605 |
},
|
1606 |
"bits": [
|
1607 |
-
|
1608 |
],
|
1609 |
"bits_prop": [
|
1610 |
1
|
@@ -1625,10 +1625,10 @@
|
|
1625 |
},
|
1626 |
"o_proj": {
|
1627 |
"group_size": {
|
1628 |
-
"
|
1629 |
},
|
1630 |
"bits": [
|
1631 |
-
|
1632 |
],
|
1633 |
"bits_prop": [
|
1634 |
1
|
@@ -1661,10 +1661,10 @@
|
|
1661 |
},
|
1662 |
"down_proj": {
|
1663 |
"group_size": {
|
1664 |
-
"
|
1665 |
},
|
1666 |
"bits": [
|
1667 |
-
|
1668 |
],
|
1669 |
"bits_prop": [
|
1670 |
1
|
@@ -1673,14 +1673,14 @@
|
|
1673 |
}
|
1674 |
},
|
1675 |
"model.layers.19": {
|
1676 |
-
"accuracy": 0.
|
1677 |
-
"total_bits":
|
1678 |
"q_proj": {
|
1679 |
"group_size": {
|
1680 |
-
"
|
1681 |
},
|
1682 |
"bits": [
|
1683 |
-
|
1684 |
],
|
1685 |
"bits_prop": [
|
1686 |
1
|
@@ -1713,10 +1713,10 @@
|
|
1713 |
},
|
1714 |
"o_proj": {
|
1715 |
"group_size": {
|
1716 |
-
"
|
1717 |
},
|
1718 |
"bits": [
|
1719 |
-
|
1720 |
],
|
1721 |
"bits_prop": [
|
1722 |
1
|
@@ -1749,10 +1749,10 @@
|
|
1749 |
},
|
1750 |
"down_proj": {
|
1751 |
"group_size": {
|
1752 |
-
"
|
1753 |
},
|
1754 |
"bits": [
|
1755 |
-
|
1756 |
],
|
1757 |
"bits_prop": [
|
1758 |
1
|
@@ -1761,8 +1761,8 @@
|
|
1761 |
}
|
1762 |
},
|
1763 |
"model.layers.20": {
|
1764 |
-
"accuracy": 0.
|
1765 |
-
"total_bits":
|
1766 |
"q_proj": {
|
1767 |
"group_size": {
|
1768 |
"2": 64
|
@@ -1801,10 +1801,10 @@
|
|
1801 |
},
|
1802 |
"o_proj": {
|
1803 |
"group_size": {
|
1804 |
-
"
|
1805 |
},
|
1806 |
"bits": [
|
1807 |
-
|
1808 |
],
|
1809 |
"bits_prop": [
|
1810 |
1
|
@@ -1813,10 +1813,10 @@
|
|
1813 |
},
|
1814 |
"up_proj": {
|
1815 |
"group_size": {
|
1816 |
-
"
|
1817 |
},
|
1818 |
"bits": [
|
1819 |
-
|
1820 |
],
|
1821 |
"bits_prop": [
|
1822 |
1
|
@@ -1837,10 +1837,10 @@
|
|
1837 |
},
|
1838 |
"down_proj": {
|
1839 |
"group_size": {
|
1840 |
-
"
|
1841 |
},
|
1842 |
"bits": [
|
1843 |
-
|
1844 |
],
|
1845 |
"bits_prop": [
|
1846 |
1
|
@@ -1849,8 +1849,8 @@
|
|
1849 |
}
|
1850 |
},
|
1851 |
"model.layers.21": {
|
1852 |
-
"accuracy": 0.
|
1853 |
-
"total_bits":
|
1854 |
"q_proj": {
|
1855 |
"group_size": {
|
1856 |
"2": 64
|
@@ -1901,10 +1901,10 @@
|
|
1901 |
},
|
1902 |
"up_proj": {
|
1903 |
"group_size": {
|
1904 |
-
"
|
1905 |
},
|
1906 |
"bits": [
|
1907 |
-
|
1908 |
],
|
1909 |
"bits_prop": [
|
1910 |
1
|
@@ -1925,10 +1925,10 @@
|
|
1925 |
},
|
1926 |
"down_proj": {
|
1927 |
"group_size": {
|
1928 |
-
"
|
1929 |
},
|
1930 |
"bits": [
|
1931 |
-
|
1932 |
],
|
1933 |
"bits_prop": [
|
1934 |
1
|
@@ -1937,8 +1937,8 @@
|
|
1937 |
}
|
1938 |
},
|
1939 |
"model.layers.22": {
|
1940 |
-
"accuracy": 0.
|
1941 |
-
"total_bits":
|
1942 |
"q_proj": {
|
1943 |
"group_size": {
|
1944 |
"2": 64
|
@@ -1953,10 +1953,10 @@
|
|
1953 |
},
|
1954 |
"k_proj": {
|
1955 |
"group_size": {
|
1956 |
-
"
|
1957 |
},
|
1958 |
"bits": [
|
1959 |
-
|
1960 |
],
|
1961 |
"bits_prop": [
|
1962 |
1
|
@@ -1977,10 +1977,10 @@
|
|
1977 |
},
|
1978 |
"o_proj": {
|
1979 |
"group_size": {
|
1980 |
-
"
|
1981 |
},
|
1982 |
"bits": [
|
1983 |
-
|
1984 |
],
|
1985 |
"bits_prop": [
|
1986 |
1
|
@@ -1989,10 +1989,10 @@
|
|
1989 |
},
|
1990 |
"up_proj": {
|
1991 |
"group_size": {
|
1992 |
-
"
|
1993 |
},
|
1994 |
"bits": [
|
1995 |
-
|
1996 |
],
|
1997 |
"bits_prop": [
|
1998 |
1
|
@@ -2013,10 +2013,10 @@
|
|
2013 |
},
|
2014 |
"down_proj": {
|
2015 |
"group_size": {
|
2016 |
-
"
|
2017 |
},
|
2018 |
"bits": [
|
2019 |
-
|
2020 |
],
|
2021 |
"bits_prop": [
|
2022 |
1
|
@@ -2025,8 +2025,8 @@
|
|
2025 |
}
|
2026 |
},
|
2027 |
"model.layers.23": {
|
2028 |
-
"accuracy": 0.
|
2029 |
-
"total_bits":
|
2030 |
"q_proj": {
|
2031 |
"group_size": {
|
2032 |
"2": 64
|
@@ -2041,10 +2041,10 @@
|
|
2041 |
},
|
2042 |
"k_proj": {
|
2043 |
"group_size": {
|
2044 |
-
"
|
2045 |
},
|
2046 |
"bits": [
|
2047 |
-
|
2048 |
],
|
2049 |
"bits_prop": [
|
2050 |
1
|
@@ -2065,10 +2065,10 @@
|
|
2065 |
},
|
2066 |
"o_proj": {
|
2067 |
"group_size": {
|
2068 |
-
"
|
2069 |
},
|
2070 |
"bits": [
|
2071 |
-
|
2072 |
],
|
2073 |
"bits_prop": [
|
2074 |
1
|
@@ -2077,10 +2077,10 @@
|
|
2077 |
},
|
2078 |
"up_proj": {
|
2079 |
"group_size": {
|
2080 |
-
"
|
2081 |
},
|
2082 |
"bits": [
|
2083 |
-
|
2084 |
],
|
2085 |
"bits_prop": [
|
2086 |
1
|
@@ -2101,10 +2101,10 @@
|
|
2101 |
},
|
2102 |
"down_proj": {
|
2103 |
"group_size": {
|
2104 |
-
"
|
2105 |
},
|
2106 |
"bits": [
|
2107 |
-
|
2108 |
],
|
2109 |
"bits_prop": [
|
2110 |
1
|
@@ -2113,8 +2113,8 @@
|
|
2113 |
}
|
2114 |
},
|
2115 |
"model.layers.24": {
|
2116 |
-
"accuracy": 0.
|
2117 |
-
"total_bits":
|
2118 |
"q_proj": {
|
2119 |
"group_size": {
|
2120 |
"2": 64
|
@@ -2129,10 +2129,10 @@
|
|
2129 |
},
|
2130 |
"k_proj": {
|
2131 |
"group_size": {
|
2132 |
-
"
|
2133 |
},
|
2134 |
"bits": [
|
2135 |
-
|
2136 |
],
|
2137 |
"bits_prop": [
|
2138 |
1
|
@@ -2153,10 +2153,10 @@
|
|
2153 |
},
|
2154 |
"o_proj": {
|
2155 |
"group_size": {
|
2156 |
-
"
|
2157 |
},
|
2158 |
"bits": [
|
2159 |
-
|
2160 |
],
|
2161 |
"bits_prop": [
|
2162 |
1
|
@@ -2165,10 +2165,10 @@
|
|
2165 |
},
|
2166 |
"up_proj": {
|
2167 |
"group_size": {
|
2168 |
-
"
|
2169 |
},
|
2170 |
"bits": [
|
2171 |
-
|
2172 |
],
|
2173 |
"bits_prop": [
|
2174 |
1
|
@@ -2189,10 +2189,10 @@
|
|
2189 |
},
|
2190 |
"down_proj": {
|
2191 |
"group_size": {
|
2192 |
-
"
|
2193 |
},
|
2194 |
"bits": [
|
2195 |
-
|
2196 |
],
|
2197 |
"bits_prop": [
|
2198 |
1
|
@@ -2201,8 +2201,8 @@
|
|
2201 |
}
|
2202 |
},
|
2203 |
"model.layers.25": {
|
2204 |
-
"accuracy": 0.
|
2205 |
-
"total_bits":
|
2206 |
"q_proj": {
|
2207 |
"group_size": {
|
2208 |
"2": 64
|
@@ -2253,10 +2253,10 @@
|
|
2253 |
},
|
2254 |
"up_proj": {
|
2255 |
"group_size": {
|
2256 |
-
"
|
2257 |
},
|
2258 |
"bits": [
|
2259 |
-
|
2260 |
],
|
2261 |
"bits_prop": [
|
2262 |
1
|
@@ -2277,10 +2277,10 @@
|
|
2277 |
},
|
2278 |
"down_proj": {
|
2279 |
"group_size": {
|
2280 |
-
"
|
2281 |
},
|
2282 |
"bits": [
|
2283 |
-
|
2284 |
],
|
2285 |
"bits_prop": [
|
2286 |
1
|
@@ -2289,8 +2289,8 @@
|
|
2289 |
}
|
2290 |
},
|
2291 |
"model.layers.26": {
|
2292 |
-
"accuracy": 0.
|
2293 |
-
"total_bits":
|
2294 |
"q_proj": {
|
2295 |
"group_size": {
|
2296 |
"2": 64
|
@@ -2329,10 +2329,10 @@
|
|
2329 |
},
|
2330 |
"o_proj": {
|
2331 |
"group_size": {
|
2332 |
-
"
|
2333 |
},
|
2334 |
"bits": [
|
2335 |
-
|
2336 |
],
|
2337 |
"bits_prop": [
|
2338 |
1
|
@@ -2341,10 +2341,10 @@
|
|
2341 |
},
|
2342 |
"up_proj": {
|
2343 |
"group_size": {
|
2344 |
-
"
|
2345 |
},
|
2346 |
"bits": [
|
2347 |
-
|
2348 |
],
|
2349 |
"bits_prop": [
|
2350 |
1
|
@@ -2377,8 +2377,8 @@
|
|
2377 |
}
|
2378 |
},
|
2379 |
"model.layers.27": {
|
2380 |
-
"accuracy": 0.
|
2381 |
-
"total_bits":
|
2382 |
"q_proj": {
|
2383 |
"group_size": {
|
2384 |
"2": 64
|
@@ -2417,10 +2417,10 @@
|
|
2417 |
},
|
2418 |
"o_proj": {
|
2419 |
"group_size": {
|
2420 |
-
"
|
2421 |
},
|
2422 |
"bits": [
|
2423 |
-
|
2424 |
],
|
2425 |
"bits_prop": [
|
2426 |
1
|
@@ -2429,10 +2429,10 @@
|
|
2429 |
},
|
2430 |
"up_proj": {
|
2431 |
"group_size": {
|
2432 |
-
"
|
2433 |
},
|
2434 |
"bits": [
|
2435 |
-
|
2436 |
],
|
2437 |
"bits_prop": [
|
2438 |
1
|
@@ -2453,10 +2453,10 @@
|
|
2453 |
},
|
2454 |
"down_proj": {
|
2455 |
"group_size": {
|
2456 |
-
"
|
2457 |
},
|
2458 |
"bits": [
|
2459 |
-
|
2460 |
],
|
2461 |
"bits_prop": [
|
2462 |
1
|
@@ -2465,8 +2465,8 @@
|
|
2465 |
}
|
2466 |
},
|
2467 |
"model.layers.28": {
|
2468 |
-
"accuracy": 0.
|
2469 |
-
"total_bits":
|
2470 |
"q_proj": {
|
2471 |
"group_size": {
|
2472 |
"2": 64
|
@@ -2517,10 +2517,10 @@
|
|
2517 |
},
|
2518 |
"up_proj": {
|
2519 |
"group_size": {
|
2520 |
-
"
|
2521 |
},
|
2522 |
"bits": [
|
2523 |
-
|
2524 |
],
|
2525 |
"bits_prop": [
|
2526 |
1
|
@@ -2541,10 +2541,10 @@
|
|
2541 |
},
|
2542 |
"down_proj": {
|
2543 |
"group_size": {
|
2544 |
-
"
|
2545 |
},
|
2546 |
"bits": [
|
2547 |
-
|
2548 |
],
|
2549 |
"bits_prop": [
|
2550 |
1
|
@@ -2553,8 +2553,8 @@
|
|
2553 |
}
|
2554 |
},
|
2555 |
"model.layers.29": {
|
2556 |
-
"accuracy": 0.
|
2557 |
-
"total_bits":
|
2558 |
"q_proj": {
|
2559 |
"group_size": {
|
2560 |
"2": 64
|
@@ -2593,10 +2593,10 @@
|
|
2593 |
},
|
2594 |
"o_proj": {
|
2595 |
"group_size": {
|
2596 |
-
"
|
2597 |
},
|
2598 |
"bits": [
|
2599 |
-
|
2600 |
],
|
2601 |
"bits_prop": [
|
2602 |
1
|
@@ -2605,10 +2605,10 @@
|
|
2605 |
},
|
2606 |
"up_proj": {
|
2607 |
"group_size": {
|
2608 |
-
"
|
2609 |
},
|
2610 |
"bits": [
|
2611 |
-
|
2612 |
],
|
2613 |
"bits_prop": [
|
2614 |
1
|
@@ -2629,10 +2629,10 @@
|
|
2629 |
},
|
2630 |
"down_proj": {
|
2631 |
"group_size": {
|
2632 |
-
"
|
2633 |
},
|
2634 |
"bits": [
|
2635 |
-
|
2636 |
],
|
2637 |
"bits_prop": [
|
2638 |
1
|
@@ -2641,8 +2641,8 @@
|
|
2641 |
}
|
2642 |
},
|
2643 |
"model.layers.30": {
|
2644 |
-
"accuracy": 0.
|
2645 |
-
"total_bits":
|
2646 |
"q_proj": {
|
2647 |
"group_size": {
|
2648 |
"2": 64
|
@@ -2693,10 +2693,10 @@
|
|
2693 |
},
|
2694 |
"up_proj": {
|
2695 |
"group_size": {
|
2696 |
-
"
|
2697 |
},
|
2698 |
"bits": [
|
2699 |
-
|
2700 |
],
|
2701 |
"bits_prop": [
|
2702 |
1
|
@@ -2717,10 +2717,10 @@
|
|
2717 |
},
|
2718 |
"down_proj": {
|
2719 |
"group_size": {
|
2720 |
-
"
|
2721 |
},
|
2722 |
"bits": [
|
2723 |
-
|
2724 |
],
|
2725 |
"bits_prop": [
|
2726 |
1
|
@@ -2729,14 +2729,14 @@
|
|
2729 |
}
|
2730 |
},
|
2731 |
"model.layers.31": {
|
2732 |
-
"accuracy": 0.
|
2733 |
-
"total_bits":
|
2734 |
"q_proj": {
|
2735 |
"group_size": {
|
2736 |
-
"
|
2737 |
},
|
2738 |
"bits": [
|
2739 |
-
|
2740 |
],
|
2741 |
"bits_prop": [
|
2742 |
1
|
@@ -2745,10 +2745,10 @@
|
|
2745 |
},
|
2746 |
"k_proj": {
|
2747 |
"group_size": {
|
2748 |
-
"
|
2749 |
},
|
2750 |
"bits": [
|
2751 |
-
|
2752 |
],
|
2753 |
"bits_prop": [
|
2754 |
1
|
@@ -2769,10 +2769,10 @@
|
|
2769 |
},
|
2770 |
"o_proj": {
|
2771 |
"group_size": {
|
2772 |
-
"
|
2773 |
},
|
2774 |
"bits": [
|
2775 |
-
|
2776 |
],
|
2777 |
"bits_prop": [
|
2778 |
1
|
@@ -2781,10 +2781,10 @@
|
|
2781 |
},
|
2782 |
"up_proj": {
|
2783 |
"group_size": {
|
2784 |
-
"
|
2785 |
},
|
2786 |
"bits": [
|
2787 |
-
|
2788 |
],
|
2789 |
"bits_prop": [
|
2790 |
1
|
@@ -2805,10 +2805,10 @@
|
|
2805 |
},
|
2806 |
"down_proj": {
|
2807 |
"group_size": {
|
2808 |
-
"
|
2809 |
},
|
2810 |
"bits": [
|
2811 |
-
|
2812 |
],
|
2813 |
"bits_prop": [
|
2814 |
1
|
|
|
1 |
{
|
2 |
"measurement": {
|
3 |
"model.layers.0": {
|
4 |
+
"accuracy": 0.9363183975219727,
|
5 |
+
"total_bits": 615009024,
|
6 |
"q_proj": {
|
7 |
"group_size": {
|
8 |
"2": 64
|
|
|
17 |
},
|
18 |
"k_proj": {
|
19 |
"group_size": {
|
20 |
+
"4": 128
|
21 |
},
|
22 |
"bits": [
|
23 |
+
4
|
24 |
],
|
25 |
"bits_prop": [
|
26 |
1
|
|
|
41 |
},
|
42 |
"o_proj": {
|
43 |
"group_size": {
|
44 |
+
"4": 128
|
45 |
},
|
46 |
"bits": [
|
47 |
+
4
|
48 |
],
|
49 |
"bits_prop": [
|
50 |
1
|
|
|
77 |
},
|
78 |
"down_proj": {
|
79 |
"group_size": {
|
80 |
+
"4": 128
|
81 |
},
|
82 |
"bits": [
|
83 |
+
4
|
84 |
],
|
85 |
"bits_prop": [
|
86 |
1
|
|
|
177 |
}
|
178 |
},
|
179 |
"model.layers.2": {
|
180 |
+
"accuracy": 0.9026820063591003,
|
181 |
+
"total_bits": 458124288,
|
182 |
"q_proj": {
|
183 |
"group_size": {
|
184 |
"2": 64
|
|
|
193 |
},
|
194 |
"k_proj": {
|
195 |
"group_size": {
|
196 |
+
"2": 64
|
197 |
},
|
198 |
"bits": [
|
199 |
+
2
|
200 |
],
|
201 |
"bits_prop": [
|
202 |
1
|
|
|
265 |
}
|
266 |
},
|
267 |
"model.layers.3": {
|
268 |
+
"accuracy": 0.9316545128822327,
|
269 |
+
"total_bits": 606753024,
|
270 |
"q_proj": {
|
271 |
"group_size": {
|
272 |
"2": 64
|
|
|
281 |
},
|
282 |
"k_proj": {
|
283 |
"group_size": {
|
284 |
+
"2": 64
|
285 |
},
|
286 |
"bits": [
|
287 |
+
2
|
288 |
],
|
289 |
"bits_prop": [
|
290 |
1
|
|
|
305 |
},
|
306 |
"o_proj": {
|
307 |
"group_size": {
|
308 |
+
"4": 128
|
309 |
},
|
310 |
"bits": [
|
311 |
+
4
|
312 |
],
|
313 |
"bits_prop": [
|
314 |
1
|
|
|
341 |
},
|
342 |
"down_proj": {
|
343 |
"group_size": {
|
344 |
+
"4": 128
|
345 |
},
|
346 |
"bits": [
|
347 |
+
4
|
348 |
],
|
349 |
"bits_prop": [
|
350 |
1
|
|
|
353 |
}
|
354 |
},
|
355 |
"model.layers.4": {
|
356 |
+
"accuracy": 0.9349029064178467,
|
357 |
+
"total_bits": 573724416,
|
358 |
"q_proj": {
|
359 |
"group_size": {
|
360 |
"2": 64
|
|
|
429 |
},
|
430 |
"down_proj": {
|
431 |
"group_size": {
|
432 |
+
"4": 128
|
433 |
},
|
434 |
"bits": [
|
435 |
+
4
|
436 |
],
|
437 |
"bits_prop": [
|
438 |
1
|
|
|
441 |
}
|
442 |
},
|
443 |
"model.layers.5": {
|
444 |
+
"accuracy": 0.915228009223938,
|
445 |
+
"total_bits": 639781632,
|
446 |
"q_proj": {
|
447 |
"group_size": {
|
448 |
+
"4": 128
|
449 |
},
|
450 |
"bits": [
|
451 |
+
4
|
452 |
],
|
453 |
"bits_prop": [
|
454 |
1
|
|
|
457 |
},
|
458 |
"k_proj": {
|
459 |
"group_size": {
|
460 |
+
"2": 64
|
461 |
},
|
462 |
"bits": [
|
463 |
+
2
|
464 |
],
|
465 |
"bits_prop": [
|
466 |
1
|
|
|
481 |
},
|
482 |
"o_proj": {
|
483 |
"group_size": {
|
484 |
+
"4": 128
|
485 |
},
|
486 |
"bits": [
|
487 |
+
4
|
488 |
],
|
489 |
"bits_prop": [
|
490 |
1
|
|
|
517 |
},
|
518 |
"down_proj": {
|
519 |
"group_size": {
|
520 |
+
"4": 128
|
521 |
},
|
522 |
"bits": [
|
523 |
+
4
|
524 |
],
|
525 |
"bits_prop": [
|
526 |
1
|
|
|
529 |
}
|
530 |
},
|
531 |
"model.layers.6": {
|
532 |
+
"accuracy": 0.9211175441741943,
|
533 |
+
"total_bits": 639781632,
|
534 |
"q_proj": {
|
535 |
"group_size": {
|
536 |
+
"4": 128
|
537 |
},
|
538 |
"bits": [
|
539 |
+
4
|
540 |
],
|
541 |
"bits_prop": [
|
542 |
1
|
|
|
545 |
},
|
546 |
"k_proj": {
|
547 |
"group_size": {
|
548 |
+
"2": 64
|
549 |
},
|
550 |
"bits": [
|
551 |
+
2
|
552 |
],
|
553 |
"bits_prop": [
|
554 |
1
|
|
|
569 |
},
|
570 |
"o_proj": {
|
571 |
"group_size": {
|
572 |
+
"4": 128
|
573 |
},
|
574 |
"bits": [
|
575 |
+
4
|
576 |
],
|
577 |
"bits_prop": [
|
578 |
1
|
|
|
605 |
},
|
606 |
"down_proj": {
|
607 |
"group_size": {
|
608 |
+
"4": 128
|
609 |
},
|
610 |
"bits": [
|
611 |
+
4
|
612 |
],
|
613 |
"bits_prop": [
|
614 |
1
|
|
|
617 |
}
|
618 |
},
|
619 |
"model.layers.7": {
|
620 |
+
"accuracy": 0.9153730074564616,
|
621 |
+
"total_bits": 615009024,
|
622 |
"q_proj": {
|
623 |
"group_size": {
|
624 |
+
"2": 64
|
625 |
},
|
626 |
"bits": [
|
627 |
+
2
|
628 |
],
|
629 |
"bits_prop": [
|
630 |
1
|
|
|
633 |
},
|
634 |
"k_proj": {
|
635 |
"group_size": {
|
636 |
+
"4": 128
|
637 |
},
|
638 |
"bits": [
|
639 |
+
4
|
640 |
],
|
641 |
"bits_prop": [
|
642 |
1
|
|
|
657 |
},
|
658 |
"o_proj": {
|
659 |
"group_size": {
|
660 |
+
"4": 128
|
661 |
},
|
662 |
"bits": [
|
663 |
+
4
|
664 |
],
|
665 |
"bits_prop": [
|
666 |
1
|
|
|
693 |
},
|
694 |
"down_proj": {
|
695 |
"group_size": {
|
696 |
+
"4": 128
|
697 |
},
|
698 |
"bits": [
|
699 |
+
4
|
700 |
],
|
701 |
"bits_prop": [
|
702 |
1
|
|
|
705 |
}
|
706 |
},
|
707 |
"model.layers.8": {
|
708 |
+
"accuracy": 0.9115104675292969,
|
709 |
+
"total_bits": 606753024,
|
710 |
"q_proj": {
|
711 |
"group_size": {
|
712 |
"2": 64
|
|
|
781 |
},
|
782 |
"down_proj": {
|
783 |
"group_size": {
|
784 |
+
"4": 128
|
785 |
},
|
786 |
"bits": [
|
787 |
+
4
|
788 |
],
|
789 |
"bits_prop": [
|
790 |
1
|
|
|
793 |
}
|
794 |
},
|
795 |
"model.layers.9": {
|
796 |
+
"accuracy": 0.9099299907684326,
|
797 |
+
"total_bits": 606753024,
|
798 |
"q_proj": {
|
799 |
"group_size": {
|
800 |
"2": 64
|
|
|
809 |
},
|
810 |
"k_proj": {
|
811 |
"group_size": {
|
812 |
+
"2": 64
|
813 |
},
|
814 |
"bits": [
|
815 |
+
2
|
816 |
],
|
817 |
"bits_prop": [
|
818 |
1
|
|
|
833 |
},
|
834 |
"o_proj": {
|
835 |
"group_size": {
|
836 |
+
"4": 128
|
837 |
},
|
838 |
"bits": [
|
839 |
+
4
|
840 |
],
|
841 |
"bits_prop": [
|
842 |
1
|
|
|
869 |
},
|
870 |
"down_proj": {
|
871 |
"group_size": {
|
872 |
+
"4": 128
|
873 |
},
|
874 |
"bits": [
|
875 |
+
4
|
876 |
],
|
877 |
"bits_prop": [
|
878 |
1
|
|
|
881 |
}
|
882 |
},
|
883 |
"model.layers.10": {
|
884 |
+
"accuracy": 0.9182359377543131,
|
885 |
+
"total_bits": 615009024,
|
886 |
"q_proj": {
|
887 |
"group_size": {
|
888 |
"2": 64
|
|
|
921 |
},
|
922 |
"o_proj": {
|
923 |
"group_size": {
|
924 |
+
"4": 128
|
925 |
},
|
926 |
"bits": [
|
927 |
+
4
|
928 |
],
|
929 |
"bits_prop": [
|
930 |
1
|
|
|
957 |
},
|
958 |
"down_proj": {
|
959 |
"group_size": {
|
960 |
+
"4": 128
|
961 |
},
|
962 |
"bits": [
|
963 |
+
4
|
964 |
],
|
965 |
"bits_prop": [
|
966 |
1
|
|
|
969 |
}
|
970 |
},
|
971 |
"model.layers.11": {
|
972 |
+
"accuracy": 0.9103440443674724,
|
973 |
+
"total_bits": 615009024,
|
974 |
"q_proj": {
|
975 |
"group_size": {
|
976 |
"2": 64
|
|
|
985 |
},
|
986 |
"k_proj": {
|
987 |
"group_size": {
|
988 |
+
"4": 128
|
989 |
},
|
990 |
"bits": [
|
991 |
+
4
|
992 |
],
|
993 |
"bits_prop": [
|
994 |
1
|
|
|
1045 |
},
|
1046 |
"down_proj": {
|
1047 |
"group_size": {
|
1048 |
+
"4": 128
|
1049 |
},
|
1050 |
"bits": [
|
1051 |
+
4
|
1052 |
],
|
1053 |
"bits_prop": [
|
1054 |
1
|
|
|
1057 |
}
|
1058 |
},
|
1059 |
"model.layers.12": {
|
1060 |
+
"accuracy": 0.9088516235351562,
|
1061 |
+
"total_bits": 639781632,
|
1062 |
"q_proj": {
|
1063 |
"group_size": {
|
1064 |
+
"4": 128
|
1065 |
},
|
1066 |
"bits": [
|
1067 |
+
4
|
1068 |
],
|
1069 |
"bits_prop": [
|
1070 |
1
|
|
|
1073 |
},
|
1074 |
"k_proj": {
|
1075 |
"group_size": {
|
1076 |
+
"2": 64
|
1077 |
},
|
1078 |
"bits": [
|
1079 |
+
2
|
1080 |
],
|
1081 |
"bits_prop": [
|
1082 |
1
|
|
|
1097 |
},
|
1098 |
"o_proj": {
|
1099 |
"group_size": {
|
1100 |
+
"4": 128
|
1101 |
},
|
1102 |
"bits": [
|
1103 |
+
4
|
1104 |
],
|
1105 |
"bits_prop": [
|
1106 |
1
|
|
|
1133 |
},
|
1134 |
"down_proj": {
|
1135 |
"group_size": {
|
1136 |
+
"4": 128
|
1137 |
},
|
1138 |
"bits": [
|
1139 |
+
4
|
1140 |
],
|
1141 |
"bits_prop": [
|
1142 |
1
|
|
|
1145 |
}
|
1146 |
},
|
1147 |
"model.layers.13": {
|
1148 |
+
"accuracy": 0.9123655954996744,
|
1149 |
+
"total_bits": 615009024,
|
1150 |
"q_proj": {
|
1151 |
"group_size": {
|
1152 |
"2": 64
|
|
|
1161 |
},
|
1162 |
"k_proj": {
|
1163 |
"group_size": {
|
1164 |
+
"4": 128
|
1165 |
},
|
1166 |
"bits": [
|
1167 |
+
4
|
1168 |
],
|
1169 |
"bits_prop": [
|
1170 |
1
|
|
|
1221 |
},
|
1222 |
"down_proj": {
|
1223 |
"group_size": {
|
1224 |
+
"4": 128
|
1225 |
},
|
1226 |
"bits": [
|
1227 |
+
4
|
1228 |
],
|
1229 |
"bits_prop": [
|
1230 |
1
|
|
|
1233 |
}
|
1234 |
},
|
1235 |
"model.layers.14": {
|
1236 |
+
"accuracy": 0.9045030275980632,
|
1237 |
+
"total_bits": 606753024,
|
1238 |
"q_proj": {
|
1239 |
"group_size": {
|
1240 |
"2": 64
|
|
|
1309 |
},
|
1310 |
"down_proj": {
|
1311 |
"group_size": {
|
1312 |
+
"4": 128
|
1313 |
},
|
1314 |
"bits": [
|
1315 |
+
4
|
1316 |
],
|
1317 |
"bits_prop": [
|
1318 |
1
|
|
|
1321 |
}
|
1322 |
},
|
1323 |
"model.layers.15": {
|
1324 |
+
"accuracy": 0.9138115247090658,
|
1325 |
+
"total_bits": 648037632,
|
1326 |
"q_proj": {
|
1327 |
"group_size": {
|
1328 |
+
"4": 128
|
1329 |
},
|
1330 |
"bits": [
|
1331 |
+
4
|
1332 |
],
|
1333 |
"bits_prop": [
|
1334 |
1
|
|
|
1337 |
},
|
1338 |
"k_proj": {
|
1339 |
"group_size": {
|
1340 |
+
"4": 128
|
1341 |
},
|
1342 |
"bits": [
|
1343 |
+
4
|
1344 |
],
|
1345 |
"bits_prop": [
|
1346 |
1
|
|
|
1397 |
},
|
1398 |
"down_proj": {
|
1399 |
"group_size": {
|
1400 |
+
"4": 128
|
1401 |
},
|
1402 |
"bits": [
|
1403 |
+
4
|
1404 |
],
|
1405 |
"bits_prop": [
|
1406 |
1
|
|
|
1409 |
}
|
1410 |
},
|
1411 |
"model.layers.16": {
|
1412 |
+
"accuracy": 0.907196839650472,
|
1413 |
+
"total_bits": 648037632,
|
1414 |
"q_proj": {
|
1415 |
"group_size": {
|
1416 |
+
"4": 128
|
1417 |
},
|
1418 |
"bits": [
|
1419 |
+
4
|
1420 |
],
|
1421 |
"bits_prop": [
|
1422 |
1
|
|
|
1425 |
},
|
1426 |
"k_proj": {
|
1427 |
"group_size": {
|
1428 |
+
"4": 128
|
1429 |
},
|
1430 |
"bits": [
|
1431 |
+
4
|
1432 |
],
|
1433 |
"bits_prop": [
|
1434 |
1
|
|
|
1485 |
},
|
1486 |
"down_proj": {
|
1487 |
"group_size": {
|
1488 |
+
"4": 128
|
1489 |
},
|
1490 |
"bits": [
|
1491 |
+
4
|
1492 |
],
|
1493 |
"bits_prop": [
|
1494 |
1
|
|
|
1497 |
}
|
1498 |
},
|
1499 |
"model.layers.17": {
|
1500 |
+
"accuracy": 0.9050076802571615,
|
1501 |
+
"total_bits": 615009024,
|
1502 |
"q_proj": {
|
1503 |
"group_size": {
|
1504 |
"2": 64
|
|
|
1513 |
},
|
1514 |
"k_proj": {
|
1515 |
"group_size": {
|
1516 |
+
"4": 128
|
1517 |
},
|
1518 |
"bits": [
|
1519 |
+
4
|
1520 |
],
|
1521 |
"bits_prop": [
|
1522 |
1
|
|
|
1537 |
},
|
1538 |
"o_proj": {
|
1539 |
"group_size": {
|
1540 |
+
"4": 128
|
1541 |
},
|
1542 |
"bits": [
|
1543 |
+
4
|
1544 |
],
|
1545 |
"bits_prop": [
|
1546 |
1
|
|
|
1573 |
},
|
1574 |
"down_proj": {
|
1575 |
"group_size": {
|
1576 |
+
"4": 128
|
1577 |
},
|
1578 |
"bits": [
|
1579 |
+
4
|
1580 |
],
|
1581 |
"bits_prop": [
|
1582 |
1
|
|
|
1585 |
}
|
1586 |
},
|
1587 |
"model.layers.18": {
|
1588 |
+
"accuracy": 0.9070329666137695,
|
1589 |
+
"total_bits": 648037632,
|
1590 |
"q_proj": {
|
1591 |
"group_size": {
|
1592 |
+
"4": 128
|
1593 |
},
|
1594 |
"bits": [
|
1595 |
+
4
|
1596 |
],
|
1597 |
"bits_prop": [
|
1598 |
1
|
|
|
1601 |
},
|
1602 |
"k_proj": {
|
1603 |
"group_size": {
|
1604 |
+
"4": 128
|
1605 |
},
|
1606 |
"bits": [
|
1607 |
+
4
|
1608 |
],
|
1609 |
"bits_prop": [
|
1610 |
1
|
|
|
1625 |
},
|
1626 |
"o_proj": {
|
1627 |
"group_size": {
|
1628 |
+
"4": 128
|
1629 |
},
|
1630 |
"bits": [
|
1631 |
+
4
|
1632 |
],
|
1633 |
"bits_prop": [
|
1634 |
1
|
|
|
1661 |
},
|
1662 |
"down_proj": {
|
1663 |
"group_size": {
|
1664 |
+
"4": 128
|
1665 |
},
|
1666 |
"bits": [
|
1667 |
+
4
|
1668 |
],
|
1669 |
"bits_prop": [
|
1670 |
1
|
|
|
1673 |
}
|
1674 |
},
|
1675 |
"model.layers.19": {
|
1676 |
+
"accuracy": 0.9032748540242513,
|
1677 |
+
"total_bits": 648037632,
|
1678 |
"q_proj": {
|
1679 |
"group_size": {
|
1680 |
+
"4": 128
|
1681 |
},
|
1682 |
"bits": [
|
1683 |
+
4
|
1684 |
],
|
1685 |
"bits_prop": [
|
1686 |
1
|
|
|
1713 |
},
|
1714 |
"o_proj": {
|
1715 |
"group_size": {
|
1716 |
+
"4": 128
|
1717 |
},
|
1718 |
"bits": [
|
1719 |
+
4
|
1720 |
],
|
1721 |
"bits_prop": [
|
1722 |
1
|
|
|
1749 |
},
|
1750 |
"down_proj": {
|
1751 |
"group_size": {
|
1752 |
+
"4": 128
|
1753 |
},
|
1754 |
"bits": [
|
1755 |
+
4
|
1756 |
],
|
1757 |
"bits_prop": [
|
1758 |
1
|
|
|
1761 |
}
|
1762 |
},
|
1763 |
"model.layers.20": {
|
1764 |
+
"accuracy": 0.9428606033325195,
|
1765 |
+
"total_bits": 722356992,
|
1766 |
"q_proj": {
|
1767 |
"group_size": {
|
1768 |
"2": 64
|
|
|
1801 |
},
|
1802 |
"o_proj": {
|
1803 |
"group_size": {
|
1804 |
+
"4": 128
|
1805 |
},
|
1806 |
"bits": [
|
1807 |
+
4
|
1808 |
],
|
1809 |
"bits_prop": [
|
1810 |
1
|
|
|
1813 |
},
|
1814 |
"up_proj": {
|
1815 |
"group_size": {
|
1816 |
+
"4": 128
|
1817 |
},
|
1818 |
"bits": [
|
1819 |
+
4
|
1820 |
],
|
1821 |
"bits_prop": [
|
1822 |
1
|
|
|
1837 |
},
|
1838 |
"down_proj": {
|
1839 |
"group_size": {
|
1840 |
+
"4": 128
|
1841 |
},
|
1842 |
"bits": [
|
1843 |
+
4
|
1844 |
],
|
1845 |
"bits_prop": [
|
1846 |
1
|
|
|
1849 |
}
|
1850 |
},
|
1851 |
"model.layers.21": {
|
1852 |
+
"accuracy": 0.9365962346394857,
|
1853 |
+
"total_bits": 722356992,
|
1854 |
"q_proj": {
|
1855 |
"group_size": {
|
1856 |
"2": 64
|
|
|
1901 |
},
|
1902 |
"up_proj": {
|
1903 |
"group_size": {
|
1904 |
+
"4": 128
|
1905 |
},
|
1906 |
"bits": [
|
1907 |
+
4
|
1908 |
],
|
1909 |
"bits_prop": [
|
1910 |
1
|
|
|
1925 |
},
|
1926 |
"down_proj": {
|
1927 |
"group_size": {
|
1928 |
+
"4": 128
|
1929 |
},
|
1930 |
"bits": [
|
1931 |
+
4
|
1932 |
],
|
1933 |
"bits_prop": [
|
1934 |
1
|
|
|
1937 |
}
|
1938 |
},
|
1939 |
"model.layers.22": {
|
1940 |
+
"accuracy": 0.9405740102132162,
|
1941 |
+
"total_bits": 722356992,
|
1942 |
"q_proj": {
|
1943 |
"group_size": {
|
1944 |
"2": 64
|
|
|
1953 |
},
|
1954 |
"k_proj": {
|
1955 |
"group_size": {
|
1956 |
+
"2": 64
|
1957 |
},
|
1958 |
"bits": [
|
1959 |
+
2
|
1960 |
],
|
1961 |
"bits_prop": [
|
1962 |
1
|
|
|
1977 |
},
|
1978 |
"o_proj": {
|
1979 |
"group_size": {
|
1980 |
+
"4": 128
|
1981 |
},
|
1982 |
"bits": [
|
1983 |
+
4
|
1984 |
],
|
1985 |
"bits_prop": [
|
1986 |
1
|
|
|
1989 |
},
|
1990 |
"up_proj": {
|
1991 |
"group_size": {
|
1992 |
+
"4": 128
|
1993 |
},
|
1994 |
"bits": [
|
1995 |
+
4
|
1996 |
],
|
1997 |
"bits_prop": [
|
1998 |
1
|
|
|
2013 |
},
|
2014 |
"down_proj": {
|
2015 |
"group_size": {
|
2016 |
+
"4": 128
|
2017 |
},
|
2018 |
"bits": [
|
2019 |
+
4
|
2020 |
],
|
2021 |
"bits_prop": [
|
2022 |
1
|
|
|
2025 |
}
|
2026 |
},
|
2027 |
"model.layers.23": {
|
2028 |
+
"accuracy": 0.9272220929463705,
|
2029 |
+
"total_bits": 722356992,
|
2030 |
"q_proj": {
|
2031 |
"group_size": {
|
2032 |
"2": 64
|
|
|
2041 |
},
|
2042 |
"k_proj": {
|
2043 |
"group_size": {
|
2044 |
+
"2": 64
|
2045 |
},
|
2046 |
"bits": [
|
2047 |
+
2
|
2048 |
],
|
2049 |
"bits_prop": [
|
2050 |
1
|
|
|
2065 |
},
|
2066 |
"o_proj": {
|
2067 |
"group_size": {
|
2068 |
+
"4": 128
|
2069 |
},
|
2070 |
"bits": [
|
2071 |
+
4
|
2072 |
],
|
2073 |
"bits_prop": [
|
2074 |
1
|
|
|
2077 |
},
|
2078 |
"up_proj": {
|
2079 |
"group_size": {
|
2080 |
+
"4": 128
|
2081 |
},
|
2082 |
"bits": [
|
2083 |
+
4
|
2084 |
],
|
2085 |
"bits_prop": [
|
2086 |
1
|
|
|
2101 |
},
|
2102 |
"down_proj": {
|
2103 |
"group_size": {
|
2104 |
+
"4": 128
|
2105 |
},
|
2106 |
"bits": [
|
2107 |
+
4
|
2108 |
],
|
2109 |
"bits_prop": [
|
2110 |
1
|
|
|
2113 |
}
|
2114 |
},
|
2115 |
"model.layers.24": {
|
2116 |
+
"accuracy": 0.9356050491333008,
|
2117 |
+
"total_bits": 722356992,
|
2118 |
"q_proj": {
|
2119 |
"group_size": {
|
2120 |
"2": 64
|
|
|
2129 |
},
|
2130 |
"k_proj": {
|
2131 |
"group_size": {
|
2132 |
+
"2": 64
|
2133 |
},
|
2134 |
"bits": [
|
2135 |
+
2
|
2136 |
],
|
2137 |
"bits_prop": [
|
2138 |
1
|
|
|
2153 |
},
|
2154 |
"o_proj": {
|
2155 |
"group_size": {
|
2156 |
+
"4": 128
|
2157 |
},
|
2158 |
"bits": [
|
2159 |
+
4
|
2160 |
],
|
2161 |
"bits_prop": [
|
2162 |
1
|
|
|
2165 |
},
|
2166 |
"up_proj": {
|
2167 |
"group_size": {
|
2168 |
+
"4": 128
|
2169 |
},
|
2170 |
"bits": [
|
2171 |
+
4
|
2172 |
],
|
2173 |
"bits_prop": [
|
2174 |
1
|
|
|
2189 |
},
|
2190 |
"down_proj": {
|
2191 |
"group_size": {
|
2192 |
+
"4": 128
|
2193 |
},
|
2194 |
"bits": [
|
2195 |
+
4
|
2196 |
],
|
2197 |
"bits_prop": [
|
2198 |
1
|
|
|
2201 |
}
|
2202 |
},
|
2203 |
"model.layers.25": {
|
2204 |
+
"accuracy": 0.9346478780110677,
|
2205 |
+
"total_bits": 722356992,
|
2206 |
"q_proj": {
|
2207 |
"group_size": {
|
2208 |
"2": 64
|
|
|
2253 |
},
|
2254 |
"up_proj": {
|
2255 |
"group_size": {
|
2256 |
+
"4": 128
|
2257 |
},
|
2258 |
"bits": [
|
2259 |
+
4
|
2260 |
],
|
2261 |
"bits_prop": [
|
2262 |
1
|
|
|
2277 |
},
|
2278 |
"down_proj": {
|
2279 |
"group_size": {
|
2280 |
+
"4": 128
|
2281 |
},
|
2282 |
"bits": [
|
2283 |
+
4
|
2284 |
],
|
2285 |
"bits_prop": [
|
2286 |
1
|
|
|
2289 |
}
|
2290 |
},
|
2291 |
"model.layers.26": {
|
2292 |
+
"accuracy": 0.9298241933186849,
|
2293 |
+
"total_bits": 722356992,
|
2294 |
"q_proj": {
|
2295 |
"group_size": {
|
2296 |
"2": 64
|
|
|
2329 |
},
|
2330 |
"o_proj": {
|
2331 |
"group_size": {
|
2332 |
+
"4": 128
|
2333 |
},
|
2334 |
"bits": [
|
2335 |
+
4
|
2336 |
],
|
2337 |
"bits_prop": [
|
2338 |
1
|
|
|
2341 |
},
|
2342 |
"up_proj": {
|
2343 |
"group_size": {
|
2344 |
+
"4": 128
|
2345 |
},
|
2346 |
"bits": [
|
2347 |
+
4
|
2348 |
],
|
2349 |
"bits_prop": [
|
2350 |
1
|
|
|
2377 |
}
|
2378 |
},
|
2379 |
"model.layers.27": {
|
2380 |
+
"accuracy": 0.9336029688517252,
|
2381 |
+
"total_bits": 722356992,
|
2382 |
"q_proj": {
|
2383 |
"group_size": {
|
2384 |
"2": 64
|
|
|
2417 |
},
|
2418 |
"o_proj": {
|
2419 |
"group_size": {
|
2420 |
+
"4": 128
|
2421 |
},
|
2422 |
"bits": [
|
2423 |
+
4
|
2424 |
],
|
2425 |
"bits_prop": [
|
2426 |
1
|
|
|
2429 |
},
|
2430 |
"up_proj": {
|
2431 |
"group_size": {
|
2432 |
+
"4": 128
|
2433 |
},
|
2434 |
"bits": [
|
2435 |
+
4
|
2436 |
],
|
2437 |
"bits_prop": [
|
2438 |
1
|
|
|
2453 |
},
|
2454 |
"down_proj": {
|
2455 |
"group_size": {
|
2456 |
+
"4": 128
|
2457 |
},
|
2458 |
"bits": [
|
2459 |
+
4
|
2460 |
],
|
2461 |
"bits_prop": [
|
2462 |
1
|
|
|
2465 |
}
|
2466 |
},
|
2467 |
"model.layers.28": {
|
2468 |
+
"accuracy": 0.9313348134358723,
|
2469 |
+
"total_bits": 722356992,
|
2470 |
"q_proj": {
|
2471 |
"group_size": {
|
2472 |
"2": 64
|
|
|
2517 |
},
|
2518 |
"up_proj": {
|
2519 |
"group_size": {
|
2520 |
+
"4": 128
|
2521 |
},
|
2522 |
"bits": [
|
2523 |
+
4
|
2524 |
],
|
2525 |
"bits_prop": [
|
2526 |
1
|
|
|
2541 |
},
|
2542 |
"down_proj": {
|
2543 |
"group_size": {
|
2544 |
+
"4": 128
|
2545 |
},
|
2546 |
"bits": [
|
2547 |
+
4
|
2548 |
],
|
2549 |
"bits_prop": [
|
2550 |
1
|
|
|
2553 |
}
|
2554 |
},
|
2555 |
"model.layers.29": {
|
2556 |
+
"accuracy": 0.93511962890625,
|
2557 |
+
"total_bits": 722356992,
|
2558 |
"q_proj": {
|
2559 |
"group_size": {
|
2560 |
"2": 64
|
|
|
2593 |
},
|
2594 |
"o_proj": {
|
2595 |
"group_size": {
|
2596 |
+
"4": 128
|
2597 |
},
|
2598 |
"bits": [
|
2599 |
+
4
|
2600 |
],
|
2601 |
"bits_prop": [
|
2602 |
1
|
|
|
2605 |
},
|
2606 |
"up_proj": {
|
2607 |
"group_size": {
|
2608 |
+
"4": 128
|
2609 |
},
|
2610 |
"bits": [
|
2611 |
+
4
|
2612 |
],
|
2613 |
"bits_prop": [
|
2614 |
1
|
|
|
2629 |
},
|
2630 |
"down_proj": {
|
2631 |
"group_size": {
|
2632 |
+
"4": 128
|
2633 |
},
|
2634 |
"bits": [
|
2635 |
+
4
|
2636 |
],
|
2637 |
"bits_prop": [
|
2638 |
1
|
|
|
2641 |
}
|
2642 |
},
|
2643 |
"model.layers.30": {
|
2644 |
+
"accuracy": 0.9314858913421631,
|
2645 |
+
"total_bits": 722356992,
|
2646 |
"q_proj": {
|
2647 |
"group_size": {
|
2648 |
"2": 64
|
|
|
2693 |
},
|
2694 |
"up_proj": {
|
2695 |
"group_size": {
|
2696 |
+
"4": 128
|
2697 |
},
|
2698 |
"bits": [
|
2699 |
+
4
|
2700 |
],
|
2701 |
"bits_prop": [
|
2702 |
1
|
|
|
2717 |
},
|
2718 |
"down_proj": {
|
2719 |
"group_size": {
|
2720 |
+
"4": 128
|
2721 |
},
|
2722 |
"bits": [
|
2723 |
+
4
|
2724 |
],
|
2725 |
"bits_prop": [
|
2726 |
1
|
|
|
2729 |
}
|
2730 |
},
|
2731 |
"model.layers.31": {
|
2732 |
+
"accuracy": 0.9324929714202881,
|
2733 |
+
"total_bits": 755385600,
|
2734 |
"q_proj": {
|
2735 |
"group_size": {
|
2736 |
+
"4": 128
|
2737 |
},
|
2738 |
"bits": [
|
2739 |
+
4
|
2740 |
],
|
2741 |
"bits_prop": [
|
2742 |
1
|
|
|
2745 |
},
|
2746 |
"k_proj": {
|
2747 |
"group_size": {
|
2748 |
+
"2": 64
|
2749 |
},
|
2750 |
"bits": [
|
2751 |
+
2
|
2752 |
],
|
2753 |
"bits_prop": [
|
2754 |
1
|
|
|
2769 |
},
|
2770 |
"o_proj": {
|
2771 |
"group_size": {
|
2772 |
+
"4": 128
|
2773 |
},
|
2774 |
"bits": [
|
2775 |
+
4
|
2776 |
],
|
2777 |
"bits_prop": [
|
2778 |
1
|
|
|
2781 |
},
|
2782 |
"up_proj": {
|
2783 |
"group_size": {
|
2784 |
+
"4": 128
|
2785 |
},
|
2786 |
"bits": [
|
2787 |
+
4
|
2788 |
],
|
2789 |
"bits_prop": [
|
2790 |
1
|
|
|
2805 |
},
|
2806 |
"down_proj": {
|
2807 |
"group_size": {
|
2808 |
+
"4": 128
|
2809 |
},
|
2810 |
"bits": [
|
2811 |
+
4
|
2812 |
],
|
2813 |
"bits_prop": [
|
2814 |
1
|