Better tables (#8)
Browse files- buster/data/document_embeddings.csv +0 -0
- buster/data/documents.csv +122 -646
- buster/docparser.py +18 -5
buster/data/document_embeddings.csv
CHANGED
The diff for this file is too large to render.
See raw diff
|
|
buster/data/documents.csv
CHANGED
@@ -95,7 +95,8 @@ framework outside of the scope of the workload manager.
|
|
95 |
If this all seems complicated, you should know that all these things
|
96 |
do not need to always be used. It is perfectly acceptable to sumbit
|
97 |
jobs with a single step, a single task and a single process.
|
98 |
-
The available
|
|
|
99 |
workload manager’s job to allocate them. Whenever a job request comes
|
100 |
in and there are not enough resources available to start it
|
101 |
immediately, it will go in the queue.
|
@@ -110,8 +111,7 @@ can see the status of your queued jobs and why they remain in the
|
|
110 |
queue.
|
111 |
The workload manager will divide the cluster into partitions according
|
112 |
to the configuration set by the admins. A partition is a set of
|
113 |
-
|
114 |
-
The workload manager,https://docs.mila.quebec/Theory_cluster.html#the-workload-manager,"nes typically reserved for a particular purpose. An example might
|
115 |
be CPU-only machines for preprocessing setup as a separate partition.
|
116 |
It is possible for multiple partitions to share resources.
|
117 |
There will always be at least one partition that is the default
|
@@ -125,7 +125,8 @@ clusters where different hardware is mixed in and not all software is
|
|
125 |
compatible with all of it (for example x86 and POWER cpus).
|
126 |
To ensure a fair share of the computing resources for all, the workload
|
127 |
manager establishes limits on the amount of resources that a single
|
128 |
-
user can
|
|
|
129 |
jobs when you go over or soft limits which will let you run jobs, but
|
130 |
only until some other job needs the resources.
|
131 |
Admin policy will determine what those exact limits are for a
|
@@ -535,7 +536,8 @@ simultaneously, it is a weighting factor of the workload manager to balance
|
|
535 |
jobs. For instance, even though we are allocated 400 GPU-years across all
|
536 |
clusters, we can use more or less than 400 GPUs simultaneously depending on the
|
537 |
history of usage from our group and other groups using the cluster at a given
|
538 |
-
period of time. Please see the Alliance’s
|
|
|
539 |
more information on how allocations and resource scheduling are configured for
|
540 |
these installations.
|
541 |
The table below provides information on the allocation for
|
@@ -543,62 +545,14 @@ rrg-bengioy-ad for the period which spans from April 2022 to
|
|
543 |
April 2023. Note that there are no special allocations for GPUs on
|
544 |
Graham and therefore jobs with GPUs should be submitted with the
|
545 |
account def-bengioy.
|
546 |
-
|
547 |
-
|
548 |
-
|
549 |
-
|
550 |
-
|
551 |
-
|
552 |
-
|
553 |
-
|
554 |
-
|
555 |
-
|
556 |
-
|
557 |
-
Cluster
|
558 |
-
CPUs
|
559 |
-
GPUs
|
560 |
-
|
561 |
-
#
|
562 |
-
account
|
563 |
-
Model
|
564 |
-
#
|
565 |
-
SLURM type specifier
|
566 |
-
account
|
567 |
-
|
568 |
-
Beluga
|
569 |
-
238
|
570 |
-
rrg-bengioy-ad
|
571 |
-
V100-16G
|
572 |
-
77
|
573 |
-
v100
|
574 |
-
rrg-bengioy-ad
|
575 |
-
|
576 |
-
Cedar
|
577 |
-
34
|
578 |
-
rrg-bengioy-ad
|
579 |
-
V100-32G
|
580 |
-
138
|
581 |
-
v100l
|
582 |
-
rrg-bengioy-ad
|
583 |
-
|
584 |
-
Graham
|
585 |
-
34
|
586 |
-
rrg-bengioy-ad
|
587 |
-
various
|
588 |
-
–
|
589 |
-
–
|
590 |
-
def-bengioy
|
591 |
-
|
592 |
-
Narval
|
593 |
-
34
|
594 |
-
rrg-bengioy-ad
|
595 |
-
A100-40G
|
596 |
-
185
|
597 |
-
a100
|
598 |
-
rrg-bengioy-ad
|
599 |
-
|
600 |
-
|
601 |
-
|
602 |
"
|
603 |
Account Creation,https://docs.mila.quebec/Extra_compute.html#account-creation,"Account Creation
|
604 |
To access the Alliance clusters you have to first create an account at
|
@@ -685,52 +639,12 @@ more time to get scheduled.
|
|
685 |
|
686 |
"
|
687 |
Beluga Storage,https://docs.mila.quebec/Extra_compute.html#beluga-storage,"Beluga Storage
|
688 |
-
|
689 |
-
|
690 |
-
|
691 |
-
|
692 |
-
|
693 |
-
|
694 |
-
|
695 |
-
Storage
|
696 |
-
Path
|
697 |
-
Usage
|
698 |
-
|
699 |
-
|
700 |
-
|
701 |
-
$HOME
|
702 |
-
/home/<user>/
|
703 |
-
|
704 |
-
Code
|
705 |
-
Specific libraries
|
706 |
-
|
707 |
-
|
708 |
-
|
709 |
-
$HOME/projects
|
710 |
-
/project/rpp-bengioy
|
711 |
-
|
712 |
-
Compressed raw datasets
|
713 |
-
|
714 |
-
|
715 |
-
|
716 |
-
$SCRATCH
|
717 |
-
/scratch/<user>
|
718 |
-
|
719 |
-
Processed datasets
|
720 |
-
Experimental results
|
721 |
-
Logs of experiments
|
722 |
-
|
723 |
-
|
724 |
-
|
725 |
-
$SLURM_TMPDIR
|
726 |
-
|
727 |
-
|
728 |
-
Temporary job results
|
729 |
-
|
730 |
-
|
731 |
-
|
732 |
-
|
733 |
-
|
734 |
They are roughly listed in order of increasing performance and optimized for
|
735 |
different uses:
|
736 |
|
@@ -758,23 +672,11 @@ Modules,https://docs.mila.quebec/Extra_compute.html#modules,"Modules
|
|
758 |
Many software, such as Python or MATLAB are already compiled and available on
|
759 |
Beluga through the module command and its subcommands. Its full
|
760 |
documentation can be found here.
|
761 |
-
|
762 |
-
|
763 |
-
|
764 |
-
|
765 |
-
|
766 |
-
|
767 |
-
module avail
|
768 |
-
Displays all the available modules
|
769 |
-
|
770 |
-
module load <module>
|
771 |
-
Loads <module>
|
772 |
-
|
773 |
-
module spider <module>
|
774 |
-
Shows specific details about <module>
|
775 |
-
|
776 |
-
|
777 |
-
|
778 |
In particular, if you with to use Python 3.6 you can simply do:
|
779 |
module load python/3.6
|
780 |
|
@@ -927,213 +829,27 @@ request them for a very short duration (for testing code before queueing long
|
|
927 |
jobs). You do not get the same guarantee as on the Mila cluster, however.
|
928 |
"
|
929 |
Node profile description,https://docs.mila.quebec/Information.html#node-profile-description,"Node profile description
|
930 |
-
|
931 |
-
|
932 |
-
|
933 |
-
|
934 |
-
|
935 |
-
|
936 |
-
|
937 |
-
|
938 |
-
|
939 |
-
|
940 |
-
|
941 |
-
|
942 |
-
|
943 |
-
|
944 |
-
|
945 |
-
|
946 |
-
|
947 |
-
GPU
|
948 |
-
|
949 |
-
|
950 |
-
|
951 |
-
Threads/Core
|
952 |
-
Memory (GB)
|
953 |
-
TmpDisk (TB)
|
954 |
-
Arch
|
955 |
-
Slurm Features
|
956 |
-
|
957 |
-
Model
|
958 |
-
Mem
|
959 |
-
#
|
960 |
-
GPU Arch and Memory
|
961 |
-
|
962 |
-
|
963 |
-
|
964 |
-
GPU Compute Nodes
|
965 |
-
|
966 |
-
cn-a[001-011]
|
967 |
-
RTX8000
|
968 |
-
48
|
969 |
-
8
|
970 |
-
40
|
971 |
-
2
|
972 |
-
20
|
973 |
-
1
|
974 |
-
384
|
975 |
-
3.6
|
976 |
-
x86_64
|
977 |
-
turing,48gb
|
978 |
-
|
979 |
-
cn-b[001-005]
|
980 |
-
V100
|
981 |
-
32
|
982 |
-
8
|
983 |
-
40
|
984 |
-
2
|
985 |
-
20
|
986 |
-
1
|
987 |
-
384
|
988 |
-
3.6
|
989 |
-
x86_64
|
990 |
-
volta,nvlink,32gb
|
991 |
-
|
992 |
-
cn-c[001-040]
|
993 |
-
RTX8000
|
994 |
-
48
|
995 |
-
8
|
996 |
-
64
|
997 |
-
2
|
998 |
-
32
|
999 |
-
1
|
1000 |
-
384
|
1001 |
-
3
|
1002 |
-
x86_64
|
1003 |
-
turing,48gb
|
1004 |
-
|
1005 |
-
cn-g[001-026]
|
1006 |
-
A100
|
1007 |
-
80
|
1008 |
-
4
|
1009 |
-
64
|
1010 |
-
2
|
1011 |
-
32
|
1012 |
-
1
|
1013 |
-
1024
|
1014 |
-
7
|
1015 |
-
x86_64
|
1016 |
-
ampere,nvlink,80gb
|
1017 |
-
|
1018 |
-
DGX Systems
|
1019 |
-
|
1020 |
-
cn-d[001-002]
|
1021 |
-
A100
|
1022 |
-
40
|
1023 |
-
8
|
1024 |
-
128
|
1025 |
-
2
|
1026 |
-
64
|
1027 |
-
1
|
1028 |
-
1024
|
1029 |
-
14
|
1030 |
-
x86_64
|
1031 |
-
ampere,nvlink,40gb
|
1032 |
-
|
1033 |
-
cn-d[003-004]
|
1034 |
-
A100
|
1035 |
-
80
|
1036 |
-
8
|
1037 |
-
128
|
1038 |
-
2
|
1039 |
-
64
|
1040 |
-
1
|
1041 |
-
2048
|
1042 |
-
28
|
1043 |
-
x86_64
|
1044 |
-
ampere,nvlink,80gb
|
1045 |
-
|
1046 |
-
cn-e[002-003]
|
1047 |
-
V100
|
1048 |
-
32
|
1049 |
-
8
|
1050 |
-
40
|
1051 |
-
2
|
1052 |
-
20
|
1053 |
-
1
|
1054 |
-
512
|
1055 |
-
7
|
1056 |
-
x86_64
|
1057 |
-
volta,32gb
|
1058 |
-
|
1059 |
-
CPU Compute Nodes
|
1060 |
-
|
1061 |
-
cn-f[001-004]
|
1062 |
-
|
1063 |
-
|
1064 |
-
|
1065 |
-
|
1066 |
-
|
1067 |
-
|
1068 |
-
|
1069 |
-
|
1070 |
-
|
1071 |
-
|
1072 |
-
|
1073 |
-
|
1074 |
-
32
|
1075 |
-
1
|
1076 |
-
32
|
1077 |
-
1
|
1078 |
-
256
|
1079 |
-
10
|
1080 |
-
x86_64
|
1081 |
-
rome
|
1082 |
-
|
1083 |
-
cn-h[001-004]
|
1084 |
-
|
1085 |
-
|
1086 |
-
|
1087 |
-
|
1088 |
-
|
1089 |
-
|
1090 |
-
|
1091 |
-
|
1092 |
-
|
1093 |
-
|
1094 |
-
|
1095 |
-
|
1096 |
-
64
|
1097 |
-
2
|
1098 |
-
32
|
1099 |
-
1
|
1100 |
-
768
|
1101 |
-
7
|
1102 |
-
x86_64
|
1103 |
-
milan
|
1104 |
-
|
1105 |
-
Legacy GPU Compute Nodes
|
1106 |
-
|
1107 |
-
kepler5
|
1108 |
-
V100
|
1109 |
-
16
|
1110 |
-
2
|
1111 |
-
16
|
1112 |
-
2
|
1113 |
-
4
|
1114 |
-
2
|
1115 |
-
256
|
1116 |
-
3.6
|
1117 |
-
x86_64
|
1118 |
-
volta,16gb
|
1119 |
-
|
1120 |
-
TITAN RTX
|
1121 |
-
|
1122 |
-
rtx[1,3-5,7]
|
1123 |
-
titanrtx
|
1124 |
-
24
|
1125 |
-
2
|
1126 |
-
20
|
1127 |
-
1
|
1128 |
-
10
|
1129 |
-
2
|
1130 |
-
128
|
1131 |
-
0.93
|
1132 |
-
x86_64
|
1133 |
-
turing,24gb
|
1134 |
-
|
1135 |
-
|
1136 |
-
|
1137 |
"
|
1138 |
Special nodes and outliers,https://docs.mila.quebec/Information.html#special-nodes-and-outliers,"Special nodes and outliers
|
1139 |
"
|
@@ -1161,55 +877,12 @@ expected to be used.
|
|
1161 |
The cn-g series of nodes include A100-80GB GPUs. One third have been
|
1162 |
configured to offer regular (non-MIG mode) a100l GPUs. The other two-thirds
|
1163 |
have been configured in MIG mode, and offer the following profiles:
|
1164 |
-
|
1165 |
-
|
1166 |
-
|
1167 |
-
|
1168 |
-
|
1169 |
-
|
1170 |
-
|
1171 |
-
|
1172 |
-
|
1173 |
-
Name
|
1174 |
-
GPU
|
1175 |
-
Cluster-wide
|
1176 |
-
|
1177 |
-
Model
|
1178 |
-
Memory
|
1179 |
-
Compute
|
1180 |
-
#
|
1181 |
-
|
1182 |
-
|
1183 |
-
|
1184 |
-
a100l.1g.10gb
|
1185 |
-
a100l.1
|
1186 |
-
A100
|
1187 |
-
10GB
|
1188 |
-
(1/8th)
|
1189 |
-
1/7th
|
1190 |
-
of full
|
1191 |
-
72
|
1192 |
-
|
1193 |
-
a100l.2g.20gb
|
1194 |
-
a100l.2
|
1195 |
-
A100
|
1196 |
-
20GB
|
1197 |
-
(2/8th)
|
1198 |
-
2/7th
|
1199 |
-
of full
|
1200 |
-
108
|
1201 |
-
|
1202 |
-
a100l.3g.40gb
|
1203 |
-
a100l.3
|
1204 |
-
A100
|
1205 |
-
40GB
|
1206 |
-
(4/8th)
|
1207 |
-
3/7th
|
1208 |
-
of full
|
1209 |
-
72
|
1210 |
-
|
1211 |
-
|
1212 |
-
|
1213 |
And can be requested using a SLURM flag such as --gres=gpu:a100l.1
|
1214 |
The partitioning may be revised as needs and SLURM capabilities evolve. Other
|
1215 |
MIG profiles exist and could be introduced.
|
@@ -1222,7 +895,6 @@ limit every MIG job to exactly one MIG slice and no more. Thus,
|
|
1222 |
--gres=gpu:a100l.3 will work (and request a size-3 slice of an
|
1223 |
a100l GPU) but --gres=gpu:a100l.1:3 (with :3 requesting
|
1224 |
three size-1 slices) will not.
|
1225 |
-
|
1226 |
"
|
1227 |
AMD,https://docs.mila.quebec/Information.html#amd,"AMD
|
1228 |
|
@@ -1329,7 +1001,8 @@ when you actually require only 8GB.
|
|
1329 |
|
1330 |
GPU
|
1331 |
Monitors the GPU usage using an nvidia-smi plugin for Netdata.
|
1332 |
-
Under the plugin interface, select the GPU
|
|
|
1333 |
you. You can figure this out by running echo $SLURM_JOB_GPUS on the
|
1334 |
allocated node or, if you have the job ID,
|
1335 |
scontrol show -d job YOUR_JOB_ID | grep 'GRES' and checking IDX
|
@@ -1363,99 +1036,20 @@ inspect this to diagnose certain problems.
|
|
1363 |
|
1364 |
|
1365 |
|
1366 |
-
|
1367 |
"
|
1368 |
Example with Mila dashboard,https://docs.mila.quebec/Information.html#example-with-mila-dashboard,"Example with Mila dashboard
|
1369 |
|
1370 |
"
|
1371 |
Storage,https://docs.mila.quebec/Information.html#storage,"Storage
|
1372 |
-
|
1373 |
-
|
1374 |
-
|
1375 |
-
|
1376 |
-
|
1377 |
-
|
1378 |
-
|
1379 |
-
|
1380 |
-
|
1381 |
-
|
1382 |
-
Path
|
1383 |
-
Performance
|
1384 |
-
Usage
|
1385 |
-
Quota (Space/Files)
|
1386 |
-
Backup
|
1387 |
-
Auto-cleanup
|
1388 |
-
|
1389 |
-
|
1390 |
-
|
1391 |
-
/network/datasets/
|
1392 |
-
High
|
1393 |
-
|
1394 |
-
Curated raw datasets (read only)
|
1395 |
-
|
1396 |
-
|
1397 |
-
|
1398 |
-
|
1399 |
-
|
1400 |
-
|
1401 |
-
$HOME or /home/mila/<u>/<username>/
|
1402 |
-
Low
|
1403 |
-
|
1404 |
-
Personal user space
|
1405 |
-
Specific libraries, code, binaries
|
1406 |
-
|
1407 |
-
|
1408 |
-
100GB/1000K
|
1409 |
-
Daily
|
1410 |
-
no
|
1411 |
-
|
1412 |
-
$SCRATCH or /network/scratch/<u>/<username>/
|
1413 |
-
High
|
1414 |
-
|
1415 |
-
Temporary job results
|
1416 |
-
Processed datasets
|
1417 |
-
Optimized for small Files
|
1418 |
-
|
1419 |
-
|
1420 |
-
no
|
1421 |
-
no
|
1422 |
-
90 days
|
1423 |
-
|
1424 |
-
$SLURM_TMPDIR
|
1425 |
-
Highest
|
1426 |
-
|
1427 |
-
High speed disk for temporary job
|
1428 |
-
results
|
1429 |
-
|
1430 |
-
|
1431 |
-
4TB/-
|
1432 |
-
no
|
1433 |
-
at job end
|
1434 |
-
|
1435 |
-
/network/projects/<groupname>/
|
1436 |
-
Fair
|
1437 |
-
|
1438 |
-
Shared space to facilitate
|
1439 |
-
collaboration between researchers
|
1440 |
-
Long-term project storage
|
1441 |
-
|
1442 |
-
|
1443 |
-
200GB/1000K
|
1444 |
-
Daily
|
1445 |
-
no
|
1446 |
-
|
1447 |
-
$ARCHIVE or /network/archive/<u>/<username>/
|
1448 |
-
Low
|
1449 |
-
|
1450 |
-
Long-term personal storage
|
1451 |
-
|
1452 |
-
|
1453 |
-
500GB
|
1454 |
-
no
|
1455 |
-
no
|
1456 |
-
|
1457 |
-
|
1458 |
-
|
1459 |
|
1460 |
Note
|
1461 |
The $HOME file system is backed up once a day. For any file
|
@@ -1758,34 +1352,13 @@ an allocation on multiple nodes.
|
|
1758 |
Job submission arguments,https://docs.mila.quebec/Userguide.html#job-submission-arguments,"Job submission arguments
|
1759 |
In order to accurately select the resources for your job, several arguments are
|
1760 |
available. The most important ones are:
|
1761 |
-
|
1762 |
-
|
1763 |
-
|
1764 |
-
|
1765 |
-
|
1766 |
-
|
1767 |
-
|
1768 |
-
Description
|
1769 |
-
|
1770 |
-
|
1771 |
-
|
1772 |
-
-n, –ntasks=<number>
|
1773 |
-
The number of task in your script, usually =1
|
1774 |
-
|
1775 |
-
-c, –cpus-per-task=<ncpus>
|
1776 |
-
The number of cores for each task
|
1777 |
-
|
1778 |
-
-t, –time=<time>
|
1779 |
-
Time requested for your job
|
1780 |
-
|
1781 |
-
–mem=<size[units]>
|
1782 |
-
Memory requested for all your tasks
|
1783 |
-
|
1784 |
-
–gres=<list>
|
1785 |
-
Select generic resources such as GPUs for your job: --gres=gpu:GPU_MODEL
|
1786 |
-
|
1787 |
-
|
1788 |
-
|
1789 |
|
1790 |
Tip
|
1791 |
Always consider requesting the adequate amount of resources to improve the
|
@@ -1816,65 +1389,23 @@ with a lower priority: unkillable > main > long. Once preempted, your job is
|
|
1816 |
killed without notice and is automatically re-queued on the same partition until
|
1817 |
resources are available. (To leverage a different preemption mechanism, see the
|
1818 |
Handling preemption)
|
1819 |
-
|
1820 |
-
|
1821 |
-
|
1822 |
-
|
1823 |
-
|
1824 |
-
|
1825 |
-
|
1826 |
-
|
1827 |
-
|
1828 |
-
Max Resource Usage
|
1829 |
-
Max Time
|
1830 |
-
Note
|
1831 |
-
|
1832 |
-
|
1833 |
-
|
1834 |
-
--partition=unkillable
|
1835 |
-
6 CPUs, mem=32G, 1 GPU
|
1836 |
-
2 days
|
1837 |
-
|
1838 |
-
|
1839 |
-
--partition=unkillable-cpu
|
1840 |
-
2 CPUs, mem=16G
|
1841 |
-
2 days
|
1842 |
-
CPU-only jobs
|
1843 |
-
|
1844 |
-
--partition=short-unkillable
|
1845 |
-
24 CPUs, mem=128G, 4 GPUs
|
1846 |
-
3 hours (!)
|
1847 |
-
Large but short jobs
|
1848 |
-
|
1849 |
-
--partition=main
|
1850 |
-
8 CPUs, mem=48G, 2 GPUs
|
1851 |
-
5 days
|
1852 |
-
|
1853 |
-
|
1854 |
-
--partition=main-cpu
|
1855 |
-
8 CPUs, mem=64G
|
1856 |
-
5 days
|
1857 |
-
CPU-only jobs
|
1858 |
-
|
1859 |
-
--partition=long
|
1860 |
-
no limit of resources
|
1861 |
-
7 days
|
1862 |
-
|
1863 |
-
|
1864 |
-
--partition=long-cpu
|
1865 |
-
no limit of resources
|
1866 |
-
7 days
|
1867 |
-
CPU-only jobs
|
1868 |
-
|
1869 |
-
|
1870 |
-
|
1871 |
|
1872 |
Warning
|
1873 |
Historically, before the 2022 introduction of CPU-only nodes (e.g. the cn-f
|
1874 |
series), CPU jobs ran side-by-side with the GPU jobs on GPU nodes. To prevent
|
1875 |
them obstructing any GPU job, they were always lowest-priority and preemptible.
|
1876 |
This was implemented by automatically assigning them to one of the now-obsolete
|
1877 |
-
|
|
|
1878 |
Do not use these partition names anymore. Prefer the *-cpu partition
|
1879 |
names defined above.
|
1880 |
For backwards-compatibility purposes, the legacy partition names are translated
|
@@ -1901,28 +1432,11 @@ accessed Node profile description.
|
|
1901 |
Example:
|
1902 |
To request a machine with 2 GPUs using NVLink, you can use
|
1903 |
sbatch -c 4 --gres=gpu:2 --constraint=nvlink
|
1904 |
-
|
1905 |
-
|
1906 |
-
|
1907 |
-
|
1908 |
-
|
1909 |
-
|
1910 |
-
Feature
|
1911 |
-
Particularities
|
1912 |
-
|
1913 |
-
|
1914 |
-
|
1915 |
-
12GB/16GB/24GB/32GB/48GB
|
1916 |
-
Request a specific amount of GPU memory
|
1917 |
-
|
1918 |
-
volta/turing/ampere
|
1919 |
-
Request a specific GPU architecture
|
1920 |
-
|
1921 |
-
nvlink
|
1922 |
-
Machine with GPUs using the NVLink interconnect technology
|
1923 |
-
|
1924 |
-
|
1925 |
-
|
1926 |
"
|
1927 |
Information on partitions/nodes,https://docs.mila.quebec/Userguide.html#information-on-partitions-nodes,"Information on partitions/nodes
|
1928 |
sinfo (ref.) provides most of the
|
@@ -1947,12 +1461,12 @@ node[10-15] 6 batch idle 2 246 16000 0 (null) (null)
|
|
1947 |
And to get statistics on a job running or terminated, use sacct with some of
|
1948 |
the fields you want to display
|
1949 |
sacct --format=User,JobID,Jobname,partition,state,time,start,end,elapsed,nnodes,ncpus,nodelist,workdir -u $USER
|
1950 |
-
User JobID JobName Partition State Timelimit Start End Elapsed NNodes NCPUS
|
|
|
1951 |
--------- ------------ ---------- ---------- ---------- ---------- ------------------- ------------------- ---------- -------- ---------- --------------- --------------------
|
1952 |
my_usern+ 2398 run_extra+ batch RUNNING 130-05:00+ 2019-03-27T18:33:43 Unknown 1-01:07:54 1 16 node9 /home/mila/my_usern+
|
1953 |
my_usern+ 2399 run_extra+ batch RUNNING 130-05:00+ 2019-03-26T08:51:38 Unknown 2-10:49:59 1 16 node9 /home/mila/my_usern+
|
1954 |
-
Or to get the list of all your previous jobs, use the --start=YYYY-MM-DD flag. You can check sacct(1) for further information about additional
|
1955 |
-
Information on partitions/nodes,https://docs.mila.quebec/Userguide.html#information-on-partitions-nodes,"ime formats.
|
1956 |
sacct -u $USER --start=2019-01-01
|
1957 |
scontrol (ref.) can be used to
|
1958 |
provide specific information on a job (currently running or recently terminated)
|
@@ -1966,7 +1480,8 @@ RunTime=2-10:41:57 TimeLimit=130-05:00:00 TimeMin=N/A
|
|
1966 |
SubmitTime=2019-03-26T08:47:17 EligibleTime=2019-03-26T08:49:18
|
1967 |
AccrueTime=2019-03-26T08:49:18
|
1968 |
StartTime=2019-03-26T08:51:38 EndTime=2019-08-03T13:51:38 Deadline=N/A
|
1969 |
-
PreemptTime=None
|
|
|
1970 |
LastSchedEval=2019-03-26T08:49:18
|
1971 |
Partition=slurm_partition AllocNode:Sid=login-node-1:14586
|
1972 |
ReqNodeList=(null) ExcNodeList=(null)
|
@@ -2000,8 +1515,7 @@ CfgTRES=cpu=16,mem=32000M,billing=3
|
|
2000 |
AllocTRES=cpu=16,mem=32000M
|
2001 |
CapWatts=n/a
|
2002 |
CurrentWatts=0 LowestJoules=0 ConsumedJoules=0
|
2003 |
-
ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/
|
2004 |
-
"
|
2005 |
Useful Commands,https://docs.mila.quebec/Userguide.html#useful-commands,"Useful Commands
|
2006 |
|
2007 |
sallocGet an interactive job and give you a shell. (ssh like) CPU only
|
@@ -2180,18 +1694,19 @@ module avail
|
|
2180 |
cuda/11.0 -> cudatoolkit/11.0 pytorch -> python/3.7/cuda/10.2/cudnn/7.6/pytorch/1.5.1 tensorflow -> python/3.7/tensorflow/2.2
|
2181 |
cuda/9.0 -> cudatoolkit/9.0 pytorch/1.4.0 -> python/3.7/cuda/10.2/cudnn/7.6/pytorch/1.4.0 tensorflow-cpu/1.15 -> python/3.7/tensorflow/1.15
|
2182 |
|
2183 |
-
-------------------------------------------------------------------------------------------------- /cvmfs/config.mila.quebec/modules/Core
|
|
|
2184 |
Mila (S,L) anaconda/3 (D) go/1.13.5 miniconda/2 mujoco/1.50 python/2.7 python/3.6 python/3.8 singularity/3.0.3 singularity/3.2.1 singularity/3.5.3 (D)
|
2185 |
anaconda/2 go/1.12.4 go/1.14 (D) miniconda/3 (D) mujoco/2.0 (D) python/3.5 python/3.7 (D) singularity/2.6.1 singularity/3.1.1 singularity/3.4.2
|
2186 |
|
2187 |
-
------------------------------------------------------------------------------------------------ /cvmfs/config.mila.quebec/modules/Compiler
|
2188 |
-
The module command,https://docs.mila.quebec/Userguide.html#the-module-command,"----------
|
2189 |
python/3.7/mujoco-py/2.0
|
2190 |
|
2191 |
-------------------------------------------------------------------------------------------------- /cvmfs/config.mila.quebec/modules/Cuda ---------------------------------------------------------------------------------------------------
|
2192 |
cuda/10.0/cudnn/7.3 cuda/10.0/nccl/2.4 cuda/10.1/nccl/2.4 cuda/11.0/nccl/2.7 cuda/9.0/nccl/2.4 cudatoolkit/9.0 cudatoolkit/10.1 cudnn/7.6/cuda/10.0/tensorrt/7.0
|
2193 |
cuda/10.0/cudnn/7.5 cuda/10.1/cudnn/7.5 cuda/10.2/cudnn/7.6 cuda/9.0/cudnn/7.3 cuda/9.2/cudnn/7.6 cudatoolkit/9.2 cudatoolkit/10.2 cudnn/7.6/cuda/10.1/tensorrt/7.0
|
2194 |
-
cuda/10
|
|
|
2195 |
|
2196 |
------------------------------------------------------------------------------------------------ /cvmfs/config.mila.quebec/modules/Pytorch --------------------------------------------------------------------------------------------------
|
2197 |
python/3.7/cuda/10.1/cudnn/7.6/pytorch/1.4.1 python/3.7/cuda/10.1/cudnn/7.6/pytorch/1.5.1 (D) python/3.7/cuda/10.2/cudnn/7.6/pytorch/1.5.0
|
@@ -2209,32 +1724,12 @@ module load python3.7
|
|
2209 |
"
|
2210 |
Available Software,https://docs.mila.quebec/Userguide.html#available-software,"Available Software
|
2211 |
Modules are divided in 5 main sections:
|
2212 |
-
|
2213 |
-
|
2214 |
-
|
2215 |
-
|
2216 |
-
|
2217 |
-
|
2218 |
-
Section
|
2219 |
-
Description
|
2220 |
-
|
2221 |
-
|
2222 |
-
|
2223 |
-
Core
|
2224 |
-
Base interpreter and software (Python, go, etc…)
|
2225 |
-
|
2226 |
-
Compiler
|
2227 |
-
Interpreter-dependent software (see the note below)
|
2228 |
-
|
2229 |
-
Cuda
|
2230 |
-
Toolkits, cudnn and related libraries
|
2231 |
-
|
2232 |
-
Pytorch/Tensorflow
|
2233 |
-
Pytorch/TF built with a specific Cuda/Cudnn
|
2234 |
-
version for Mila’s GPUs (see the related paragraph)
|
2235 |
-
|
2236 |
-
|
2237 |
-
|
2238 |
|
2239 |
Note
|
2240 |
Modules which are nested (../../..) usually depend on other software/module
|
@@ -2495,7 +1990,8 @@ From: tensorflow/tensorflow:latest-gpu-py3
|
|
2495 |
apt-get update
|
2496 |
apt-get install -y cmake libcupti-dev libyaml-dev wget unzip
|
2497 |
apt-get clean
|
2498 |
-
echo ""
|
|
|
2499 |
pip install tqdm
|
2500 |
echo ""Creating mount points""
|
2501 |
mkdir /dataset
|
@@ -2524,7 +2020,6 @@ Warning
|
|
2524 |
You always need to use sudo when you build a container from a
|
2525 |
recipe. As there is no access to sudo on the cluster, a personal computer or
|
2526 |
the use singularity hub is needed to build a container
|
2527 |
-
|
2528 |
"
|
2529 |
Build recipe on singularity hub,https://docs.mila.quebec/Userguide.html#build-recipe-on-singularity-hub,"Build recipe on singularity hub
|
2530 |
Singularity hub allows users to build containers from recipes directly on
|
@@ -2600,7 +2095,8 @@ From: pytorch/pytorch:1.0-cuda10.0-cudnn7-runtime
|
|
2600 |
mkdir /Gym && cd /Gym
|
2601 |
git clone https://github.com/openai/gym.git || true && \
|
2602 |
mkdir /Gym/.mujoco && cd /Gym/.mujoco
|
2603 |
-
wget https://www.roboti.us/
|
|
|
2604 |
unzip mjpro150_linux.zip && \
|
2605 |
wget https://www.roboti.us/download/mujoco200_linux.zip && \
|
2606 |
unzip mujoco200_linux.zip && \
|
@@ -2610,8 +2106,7 @@ From: pytorch/pytorch:1.0-cuda10.0-cudnn7-runtime
|
|
2610 |
export MUJOCO_PY_MJKEY_PATH=/Gym/.mujoco/mjkey.txt
|
2611 |
export MUJOCO_PY_MUJOCO_PATH=/Gym/.mujoco/mujoco150/
|
2612 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mjpro150/bin
|
2613 |
-
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym
|
2614 |
-
"Example: Recipe with OpenAI gym, MuJoCo and Miniworld",https://docs.mila.quebec/Userguide.html#example-recipe-with-openai-gym-mujoco-and-miniworld,"/.mujoco/mujoco200/bin
|
2615 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/bin
|
2616 |
cp /mjkey.txt /Gym/.mujoco/mjkey.txt
|
2617 |
# Install Python dependencies
|
@@ -2632,7 +2127,8 @@ From: pytorch/pytorch:1.0-cuda10.0-cudnn7-runtime
|
|
2632 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mjpro150/bin
|
2633 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mujoco200/bin
|
2634 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/bin
|
2635 |
-
export PATH=/Gym/gym/.tox/py3/bin:$PATH
|
|
|
2636 |
|
2637 |
%runscript
|
2638 |
exec /bin/sh ""$@""
|
@@ -2674,8 +2170,7 @@ From: tensorflow/tensorflow:latest-gpu-py3
|
|
2674 |
|
2675 |
# Download Gym and MuJoCo
|
2676 |
mkdir /Gym && cd /Gym
|
2677 |
-
git clone
|
2678 |
-
"Example: Recipe with OpenAI gym, MuJoCo and Miniworld",https://docs.mila.quebec/Userguide.html#example-recipe-with-openai-gym-mujoco-and-miniworld," https://github.com/openai/gym.git || true && \
|
2679 |
mkdir /Gym/.mujoco && cd /Gym/.mujoco
|
2680 |
wget https://www.roboti.us/download/mjpro150_linux.zip && \
|
2681 |
unzip mjpro150_linux.zip && \
|
@@ -2685,7 +2180,8 @@ From: tensorflow/tensorflow:latest-gpu-py3
|
|
2685 |
|
2686 |
# Export global environment variables
|
2687 |
export MUJOCO_PY_MJKEY_PATH=/Gym/.mujoco/mjkey.txt
|
2688 |
-
export MUJOCO_PY_MUJOCO_PATH=/Gym/.mujoco/
|
|
|
2689 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mjpro150/bin
|
2690 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mujoco200/bin
|
2691 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/bin
|
@@ -2722,8 +2218,7 @@ From: tensorflow/tensorflow:latest-gpu-py3
|
|
2722 |
|
2723 |
Keep in mind that those environment variables are sourced at runtime and not at
|
2724 |
build time. This is why, you should also define them in the %post section
|
2725 |
-
since they are required to install MuJoCo
|
2726 |
-
"
|
2727 |
Using containers on clusters,https://docs.mila.quebec/Userguide.html#using-containers-on-clusters,"Using containers on clusters
|
2728 |
"
|
2729 |
How to use containers on clusters,https://docs.mila.quebec/Userguide.html#how-to-use-containers-on-clusters,"How to use containers on clusters
|
@@ -3168,29 +2663,10 @@ It does not require any ssh tunnel or port redirection, the hub acts as a proxy
|
|
3168 |
server that will redirect you to a session as soon as it is available.
|
3169 |
It is currently available for Mila clusters and some Digital Research Alliance
|
3170 |
of Canada (Alliance) clusters.
|
3171 |
-
|
3172 |
-
|
3173 |
-
|
3174 |
-
|
3175 |
-
|
3176 |
-
|
3177 |
-
|
3178 |
-
Cluster
|
3179 |
-
Address
|
3180 |
-
Login type
|
3181 |
-
|
3182 |
-
|
3183 |
-
|
3184 |
-
Mila Local
|
3185 |
-
https://jupyterhub.server.mila.quebec
|
3186 |
-
Google Oauth
|
3187 |
-
|
3188 |
-
Alliance
|
3189 |
-
https://docs.alliancecan.ca/wiki/JupyterHub
|
3190 |
-
DRAC login
|
3191 |
-
|
3192 |
-
|
3193 |
-
|
3194 |
|
3195 |
Warning
|
3196 |
Do not forget to close the JupyterLab session! Closing the window leaves
|
@@ -3351,7 +2827,8 @@ Requesting 2 tasks per GPU
|
|
3351 |
|
3352 |
|
3353 |
--exclusive is important to specify subsequent step/srun to bind to different cpus.
|
3354 |
-
This will produce 8 output files
|
|
|
3355 |
|
3356 |
JOBID-step-0-task-0.out
|
3357 |
JOBID-step-0-task-1.out
|
@@ -3372,8 +2849,7 @@ cat JOBID-step-* | grep Tesla
|
|
3372 |
0: | 0 Tesla P100-PCIE... On | 00000000:82:00.0 Off | 0 |
|
3373 |
1: | 0 Tesla P100-PCIE... On | 00000000:82:00.0 Off | 0 |
|
3374 |
0: | 0 Tesla P100-PCIE... On | 00000000:03:00.0 Off | 0 |
|
3375 |
-
1: | 0 Tesla P100-PCIE... On | 00000000:03:00.0 Off | 0 |
|
3376 |
-
"
|
3377 |
Multiple Nodes,https://docs.mila.quebec/Userguide.html#multiple-nodes,"Multiple Nodes
|
3378 |
"
|
3379 |
Data Parallel,https://docs.mila.quebec/Userguide.html#data-parallel,"Data Parallel
|
|
|
95 |
If this all seems complicated, you should know that all these things
|
96 |
do not need to always be used. It is perfectly acceptable to sumbit
|
97 |
jobs with a single step, a single task and a single process.
|
98 |
+
The available resource"
|
99 |
+
The workload manager,https://docs.mila.quebec/Theory_cluster.html#the-workload-manager,"s on the cluster are not infinite and it is the
|
100 |
workload manager’s job to allocate them. Whenever a job request comes
|
101 |
in and there are not enough resources available to start it
|
102 |
immediately, it will go in the queue.
|
|
|
111 |
queue.
|
112 |
The workload manager will divide the cluster into partitions according
|
113 |
to the configuration set by the admins. A partition is a set of
|
114 |
+
machines typically reserved for a particular purpose. An example might
|
|
|
115 |
be CPU-only machines for preprocessing setup as a separate partition.
|
116 |
It is possible for multiple partitions to share resources.
|
117 |
There will always be at least one partition that is the default
|
|
|
125 |
compatible with all of it (for example x86 and POWER cpus).
|
126 |
To ensure a fair share of the computing resources for all, the workload
|
127 |
manager establishes limits on the amount of resources that a single
|
128 |
+
user can us"
|
129 |
+
The workload manager,https://docs.mila.quebec/Theory_cluster.html#the-workload-manager,"e at once. These can be hard limits which prevent running
|
130 |
jobs when you go over or soft limits which will let you run jobs, but
|
131 |
only until some other job needs the resources.
|
132 |
Admin policy will determine what those exact limits are for a
|
|
|
536 |
jobs. For instance, even though we are allocated 400 GPU-years across all
|
537 |
clusters, we can use more or less than 400 GPUs simultaneously depending on the
|
538 |
history of usage from our group and other groups using the cluster at a given
|
539 |
+
period of time. Please see the Alliance’s doc"
|
540 |
+
Current allocation description,https://docs.mila.quebec/Extra_compute.html#current-allocation-description,"umentation for
|
541 |
more information on how allocations and resource scheduling are configured for
|
542 |
these installations.
|
543 |
The table below provides information on the allocation for
|
|
|
545 |
April 2023. Note that there are no special allocations for GPUs on
|
546 |
Graham and therefore jobs with GPUs should be submitted with the
|
547 |
account def-bengioy.
|
548 |
+
| 0 | 1 | 2 | 3 | 4 | 5 | 6 |
|
549 |
+
|---------|------|----------------|----------|------|----------------------|----------------|
|
550 |
+
| Cluster | CPUs | CPUs | GPUs | GPUs | GPUs | GPUs |
|
551 |
+
| Cluster | # | account | Model | # | SLURM type specifier | account |
|
552 |
+
| Beluga | 238 | rrg-bengioy-ad | V100-16G | 77 | v100 | rrg-bengioy-ad |
|
553 |
+
| Cedar | 34 | rrg-bengioy-ad | V100-32G | 138 | v100l | rrg-bengioy-ad |
|
554 |
+
| Graham | 34 | rrg-bengioy-ad | various | – | – | def-bengioy |
|
555 |
+
| Narval | 34 | rrg-bengioy-ad | A100-40G | 185 | a100 | rrg-bengioy-ad |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
556 |
"
|
557 |
Account Creation,https://docs.mila.quebec/Extra_compute.html#account-creation,"Account Creation
|
558 |
To access the Alliance clusters you have to first create an account at
|
|
|
639 |
|
640 |
"
|
641 |
Beluga Storage,https://docs.mila.quebec/Extra_compute.html#beluga-storage,"Beluga Storage
|
642 |
+
| Storage | Path | Usage |
|
643 |
+
|----------------|----------------------|---------------------------------------------------------------|
|
644 |
+
| $HOME | /home/<user>/ | Code Specific libraries |
|
645 |
+
| $HOME/projects | /project/rpp-bengioy | Compressed raw datasets |
|
646 |
+
| $SCRATCH | /scratch/<user> | Processed datasets Experimental results Logs of experiments |
|
647 |
+
| $SLURM_TMPDIR | nan | Temporary job results |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
648 |
They are roughly listed in order of increasing performance and optimized for
|
649 |
different uses:
|
650 |
|
|
|
672 |
Many software, such as Python or MATLAB are already compiled and available on
|
673 |
Beluga through the module command and its subcommands. Its full
|
674 |
documentation can be found here.
|
675 |
+
| 0 | 1 |
|
676 |
+
|------------------------|---------------------------------------|
|
677 |
+
| module avail | Displays all the available modules |
|
678 |
+
| module load <module> | Loads <module> |
|
679 |
+
| module spider <module> | Shows specific details about <module> |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
680 |
In particular, if you with to use Python 3.6 you can simply do:
|
681 |
module load python/3.6
|
682 |
|
|
|
829 |
jobs). You do not get the same guarantee as on the Mila cluster, however.
|
830 |
"
|
831 |
Node profile description,https://docs.mila.quebec/Information.html#node-profile-description,"Node profile description
|
832 |
+
| ('Name', 'Name') | ('GPU', 'Model') | ('GPU', 'Mem') | ('GPU', '#') | ('CPUs', 'CPUs') | ('Sockets', 'Sockets') | ('Cores/Socket', 'Cores/Socket') | ('Threads/Core', 'Threads/Core') | ('Memory (GB)', 'Memory (GB)') | ('TmpDisk (TB)', 'TmpDisk (TB)') | ('Arch', 'Arch') | ('Slurm Features', 'GPU Arch and Memory') |
|
833 |
+
|--------------------------|--------------------------|--------------------------|--------------------------|--------------------------|--------------------------|------------------------------------|------------------------------------|----------------------------------|------------------------------------|--------------------------|---------------------------------------------|
|
834 |
+
| GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes | GPU Compute Nodes |
|
835 |
+
| cn-a[001-011] | RTX8000 | 48 | 8 | 40 | 2 | 20 | 1 | 384 | 3.6 | x86_64 | turing,48gb |
|
836 |
+
| cn-b[001-005] | V100 | 32 | 8 | 40 | 2 | 20 "
|
837 |
+
Node profile description,https://docs.mila.quebec/Information.html#node-profile-description," | 1 | 384 | 3.6 | x86_64 | volta,nvlink,32gb |
|
838 |
+
| cn-c[001-040] | RTX8000 | 48 | 8 | 64 | 2 | 32 | 1 | 384 | 3 | x86_64 | turing,48gb |
|
839 |
+
| cn-g[001-026] | A100 | 80 | 4 | 64 | 2 | 32 | 1 | 1024 | 7 | x86_64 | ampere,nvlink,80gb |
|
840 |
+
| DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems | DGX Systems |
|
841 |
+
| cn-d[001-002] | A100 | 40 | 8 | 128 | 2 | 64 | 1 | 1024 | 14 | x86_64 | ampere,nvlink,40gb "
|
842 |
+
Node profile description,https://docs.mila.quebec/Information.html#node-profile-description," |
|
843 |
+
| cn-d[003-004] | A100 | 80 | 8 | 128 | 2 | 64 | 1 | 2048 | 28 | x86_64 | ampere,nvlink,80gb |
|
844 |
+
| cn-e[002-003] | V100 | 32 | 8 | 40 | 2 | 20 | 1 | 512 | 7 | x86_64 | volta,32gb |
|
845 |
+
| CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes | CPU Compute Nodes |
|
846 |
+
| cn-f[001-004] | nan | nan | nan | 32 | 1 | 32 | 1 | 256 | 10 | x86_64 | rome |
|
847 |
+
| cn-h[001-004] | nan | nan | nan | 64 | 2 | 32 "
|
848 |
+
Node profile description,https://docs.mila.quebec/Information.html#node-profile-description," | 1 | 768 | 7 | x86_64 | milan |
|
849 |
+
| Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes | Legacy GPU Compute Nodes |
|
850 |
+
| kepler5 | V100 | 16 | 2 | 16 | 2 | 4 | 2 | 256 | 3.6 | x86_64 | volta,16gb |
|
851 |
+
| TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX | TITAN RTX |
|
852 |
+
| rtx[1,3-5,7] | titanrtx | 24 | 2 | 20 | 1 | 10 | 2 | 128 | 0.93 | x86_64 | turing,24gb |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
853 |
"
|
854 |
Special nodes and outliers,https://docs.mila.quebec/Information.html#special-nodes-and-outliers,"Special nodes and outliers
|
855 |
"
|
|
|
877 |
The cn-g series of nodes include A100-80GB GPUs. One third have been
|
878 |
configured to offer regular (non-MIG mode) a100l GPUs. The other two-thirds
|
879 |
have been configured in MIG mode, and offer the following profiles:
|
880 |
+
| ('Name', 'Name') | ('GPU', 'Model') | ('GPU', 'Memory') | ('GPU', 'Compute') | ('Cluster-wide', '#') |
|
881 |
+
|------------------------|--------------------|---------"
|
882 |
+
MIG,https://docs.mila.quebec/Information.html#mig,"------------|----------------------|-------------------------|
|
883 |
+
| a100l.1g.10gb a100l.1 | A100 | 10GB (1/8th) | 1/7th of full | 72 |
|
884 |
+
| a100l.2g.20gb a100l.2 | A100 | 20GB (2/8th) | 2/7th of full | 108 |
|
885 |
+
| a100l.3g.40gb a100l.3 | A100 | 40GB (4/8th) | 3/7th of full | 72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
886 |
And can be requested using a SLURM flag such as --gres=gpu:a100l.1
|
887 |
The partitioning may be revised as needs and SLURM capabilities evolve. Other
|
888 |
MIG profiles exist and could be introduced.
|
|
|
895 |
--gres=gpu:a100l.3 will work (and request a size-3 slice of an
|
896 |
a100l GPU) but --gres=gpu:a100l.1:3 (with :3 requesting
|
897 |
three size-1 slices) will not.
|
|
|
898 |
"
|
899 |
AMD,https://docs.mila.quebec/Information.html#amd,"AMD
|
900 |
|
|
|
1001 |
|
1002 |
GPU
|
1003 |
Monitors the GPU usage using an nvidia-smi plugin for Netdata.
|
1004 |
+
Under the plugin interface, select the GPU"
|
1005 |
+
Example watching the CPU/RAM/GPU usage,https://docs.mila.quebec/Information.html#example-watching-the-cpu-ram-gpu-usage," number which was allocated to
|
1006 |
you. You can figure this out by running echo $SLURM_JOB_GPUS on the
|
1007 |
allocated node or, if you have the job ID,
|
1008 |
scontrol show -d job YOUR_JOB_ID | grep 'GRES' and checking IDX
|
|
|
1036 |
|
1037 |
|
1038 |
|
|
|
1039 |
"
|
1040 |
Example with Mila dashboard,https://docs.mila.quebec/Information.html#example-with-mila-dashboard,"Example with Mila dashboard
|
1041 |
|
1042 |
"
|
1043 |
Storage,https://docs.mila.quebec/Information.html#storage,"Storage
|
1044 |
+
| Path | Performance | Usage | Quota (Space/Files) | Backup | Auto-cleanup |
|
1045 |
+
|------------------------------------------------|---------------|-----------------------------------------------------------------------------------------|-----------------------|----------|----------------|
|
1046 |
+
| /network/datasets/ | High | Curated raw datasets (read only) | nan | nan | nan |
|
1047 |
+
| $HOME or /home/mila/<u>/<username>/ | Low | Personal user space Specific libraries, code, binaries | 100GB/1000K | Daily | no |
|
1048 |
+
| $SCRATCH or /network/scratch/<u>/<username>/ | High | Temporary job results Processed datasets Optimized for small Files | no | no | 90 days "
|
1049 |
+
Storage,https://docs.mila.quebec/Information.html#storage," |
|
1050 |
+
| $SLURM_TMPDIR | Highest | High speed disk for temporary job results | 4TB/- | no | at job end |
|
1051 |
+
| /network/projects/<groupname>/ | Fair | Shared space to facilitate collaboration between researchers Long-term project storage | 200GB/1000K | Daily | no |
|
1052 |
+
| $ARCHIVE or /network/archive/<u>/<username>/ | Low | Long-term personal storage | 500GB | no | no |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1053 |
|
1054 |
Note
|
1055 |
The $HOME file system is backed up once a day. For any file
|
|
|
1352 |
Job submission arguments,https://docs.mila.quebec/Userguide.html#job-submission-arguments,"Job submission arguments
|
1353 |
In order to accurately select the resources for your job, several arguments are
|
1354 |
available. The most important ones are:
|
1355 |
+
| Argument | Description |
|
1356 |
+
|----------------------------|---------------------------------------------------------------------------|
|
1357 |
+
| -n, –ntasks=<number> | The number of task in your script, usually =1 |
|
1358 |
+
| -c, –cpus-per-task=<ncpus> | The number of cores for each task |
|
1359 |
+
| -t, –time=<time> | Time requested for your job |
|
1360 |
+
| –mem=<size[units]> | Memory requested for all your tasks |
|
1361 |
+
| –gres=<list> | Select generic resources such as GPUs for your job: --gres=gpu:GPU_MODEL |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1362 |
|
1363 |
Tip
|
1364 |
Always consider requesting the adequate amount of resources to improve the
|
|
|
1389 |
killed without notice and is automatically re-queued on the same partition until
|
1390 |
resources are available. (To leverage a different preemption mechanism, see the
|
1391 |
Handling preemption)
|
1392 |
+
| Flag | Max Resource Usage | Max Time | Note |
|
1393 |
+
|------------------------------|---------------------------|-------------|----------------------|
|
1394 |
+
| --partition=unkillable | 6 CPUs, mem=32G, 1 GPU | 2 days | nan |
|
1395 |
+
| --partition=unkillable-cpu | 2 CPUs, mem=16G | 2 days | CPU-only jobs |
|
1396 |
+
| --partition=short-unkillable | 24 CPUs, mem=128G, 4 GPUs | 3 hours (!) | Large but short jobs |
|
1397 |
+
| --partition=main | 8 CPUs, mem=48G, 2 GPUs | 5 days | nan |
|
1398 |
+
| --partition=main-cpu | 8 CPUs, mem=64G | 5 days | CPU-only jobs |
|
1399 |
+
| --partition=long | no limit of resources | 7 days | nan |
|
1400 |
+
| --partition=long-cpu | no limit of resources | 7 days | CPU-only jobs |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1401 |
|
1402 |
Warning
|
1403 |
Historically, before the 2022 introduction of CPU-only nodes (e.g. the cn-f
|
1404 |
series), CPU jobs ran side-by-side with the GPU jobs on GPU nodes. To prevent
|
1405 |
them obstructing any GPU job, they were always lowest-priority and preemptible.
|
1406 |
This was implemented by automatically assigning them to one of the now-obsolete
|
1407 |
+
part"
|
1408 |
+
Partitioning,https://docs.mila.quebec/Userguide.html#partitioning,"itions cpu_jobs, cpu_jobs_low or cpu_jobs_low-grace.
|
1409 |
Do not use these partition names anymore. Prefer the *-cpu partition
|
1410 |
names defined above.
|
1411 |
For backwards-compatibility purposes, the legacy partition names are translated
|
|
|
1432 |
Example:
|
1433 |
To request a machine with 2 GPUs using NVLink, you can use
|
1434 |
sbatch -c 4 --gres=gpu:2 --constraint=nvlink
|
1435 |
+
| Feature | Particularities |
|
1436 |
+
|--------------------------|------------------------------------------------------------|
|
1437 |
+
| 12GB/16GB/24GB/32GB/48GB | Request a specific amount of GPU memory |
|
1438 |
+
| volta/turing/ampere | Request a specific GPU architecture |
|
1439 |
+
| nvlink | Machine with GPUs using the NVLink interconnect technology |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1440 |
"
|
1441 |
Information on partitions/nodes,https://docs.mila.quebec/Userguide.html#information-on-partitions-nodes,"Information on partitions/nodes
|
1442 |
sinfo (ref.) provides most of the
|
|
|
1461 |
And to get statistics on a job running or terminated, use sacct with some of
|
1462 |
the fields you want to display
|
1463 |
sacct --format=User,JobID,Jobname,partition,state,time,start,end,elapsed,nnodes,ncpus,nodelist,workdir -u $USER
|
1464 |
+
User JobID JobName Partition State Timelimit Start End Elapsed NNodes NCPUS N"
|
1465 |
+
Information on partitions/nodes,https://docs.mila.quebec/Userguide.html#information-on-partitions-nodes,"odeList WorkDir
|
1466 |
--------- ------------ ---------- ---------- ---------- ---------- ------------------- ------------------- ---------- -------- ---------- --------------- --------------------
|
1467 |
my_usern+ 2398 run_extra+ batch RUNNING 130-05:00+ 2019-03-27T18:33:43 Unknown 1-01:07:54 1 16 node9 /home/mila/my_usern+
|
1468 |
my_usern+ 2399 run_extra+ batch RUNNING 130-05:00+ 2019-03-26T08:51:38 Unknown 2-10:49:59 1 16 node9 /home/mila/my_usern+
|
1469 |
+
Or to get the list of all your previous jobs, use the --start=YYYY-MM-DD flag. You can check sacct(1) for further information about additional time formats.
|
|
|
1470 |
sacct -u $USER --start=2019-01-01
|
1471 |
scontrol (ref.) can be used to
|
1472 |
provide specific information on a job (currently running or recently terminated)
|
|
|
1480 |
SubmitTime=2019-03-26T08:47:17 EligibleTime=2019-03-26T08:49:18
|
1481 |
AccrueTime=2019-03-26T08:49:18
|
1482 |
StartTime=2019-03-26T08:51:38 EndTime=2019-08-03T13:51:38 Deadline=N/A
|
1483 |
+
PreemptTime=None SuspendTim"
|
1484 |
+
Information on partitions/nodes,https://docs.mila.quebec/Userguide.html#information-on-partitions-nodes,"e=None SecsPreSuspend=0
|
1485 |
LastSchedEval=2019-03-26T08:49:18
|
1486 |
Partition=slurm_partition AllocNode:Sid=login-node-1:14586
|
1487 |
ReqNodeList=(null) ExcNodeList=(null)
|
|
|
1515 |
AllocTRES=cpu=16,mem=32000M
|
1516 |
CapWatts=n/a
|
1517 |
CurrentWatts=0 LowestJoules=0 ConsumedJoules=0
|
1518 |
+
ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/"
|
|
|
1519 |
Useful Commands,https://docs.mila.quebec/Userguide.html#useful-commands,"Useful Commands
|
1520 |
|
1521 |
sallocGet an interactive job and give you a shell. (ssh like) CPU only
|
|
|
1694 |
cuda/11.0 -> cudatoolkit/11.0 pytorch -> python/3.7/cuda/10.2/cudnn/7.6/pytorch/1.5.1 tensorflow -> python/3.7/tensorflow/2.2
|
1695 |
cuda/9.0 -> cudatoolkit/9.0 pytorch/1.4.0 -> python/3.7/cuda/10.2/cudnn/7.6/pytorch/1.4.0 tensorflow-cpu/1.15 -> python/3.7/tensorflow/1.15
|
1696 |
|
1697 |
+
-------------------------------------------------------------------------------------------------- /cvmfs/config.mila.quebec/modules/Core ---------------------------------"
|
1698 |
+
The module command,https://docs.mila.quebec/Userguide.html#the-module-command,"------------------------------------------------------------------
|
1699 |
Mila (S,L) anaconda/3 (D) go/1.13.5 miniconda/2 mujoco/1.50 python/2.7 python/3.6 python/3.8 singularity/3.0.3 singularity/3.2.1 singularity/3.5.3 (D)
|
1700 |
anaconda/2 go/1.12.4 go/1.14 (D) miniconda/3 (D) mujoco/2.0 (D) python/3.5 python/3.7 (D) singularity/2.6.1 singularity/3.1.1 singularity/3.4.2
|
1701 |
|
1702 |
+
------------------------------------------------------------------------------------------------ /cvmfs/config.mila.quebec/modules/Compiler -------------------------------------------------------------------------------------------------
|
|
|
1703 |
python/3.7/mujoco-py/2.0
|
1704 |
|
1705 |
-------------------------------------------------------------------------------------------------- /cvmfs/config.mila.quebec/modules/Cuda ---------------------------------------------------------------------------------------------------
|
1706 |
cuda/10.0/cudnn/7.3 cuda/10.0/nccl/2.4 cuda/10.1/nccl/2.4 cuda/11.0/nccl/2.7 cuda/9.0/nccl/2.4 cudatoolkit/9.0 cudatoolkit/10.1 cudnn/7.6/cuda/10.0/tensorrt/7.0
|
1707 |
cuda/10.0/cudnn/7.5 cuda/10.1/cudnn/7.5 cuda/10.2/cudnn/7.6 cuda/9.0/cudnn/7.3 cuda/9.2/cudnn/7.6 cudatoolkit/9.2 cudatoolkit/10.2 cudnn/7.6/cuda/10.1/tensorrt/7.0
|
1708 |
+
cuda/10"
|
1709 |
+
The module command,https://docs.mila.quebec/Userguide.html#the-module-command,".0/cudnn/7.6 (D) cuda/10.1/cudnn/7.6 (D) cuda/10.2/nccl/2.7 cuda/9.0/cudnn/7.5 (D) cuda/9.2/nccl/2.4 cudatoolkit/10.0 cudatoolkit/11.0 (D) cudnn/7.6/cuda/9.0/tensorrt/7.0
|
1710 |
|
1711 |
------------------------------------------------------------------------------------------------ /cvmfs/config.mila.quebec/modules/Pytorch --------------------------------------------------------------------------------------------------
|
1712 |
python/3.7/cuda/10.1/cudnn/7.6/pytorch/1.4.1 python/3.7/cuda/10.1/cudnn/7.6/pytorch/1.5.1 (D) python/3.7/cuda/10.2/cudnn/7.6/pytorch/1.5.0
|
|
|
1724 |
"
|
1725 |
Available Software,https://docs.mila.quebec/Userguide.html#available-software,"Available Software
|
1726 |
Modules are divided in 5 main sections:
|
1727 |
+
| Section | Description |
|
1728 |
+
|--------------------|-----------------------------------------------------------------------------------------------------|
|
1729 |
+
| Core | Base interpreter and software (Python, go, etc…) |
|
1730 |
+
| Compiler | Interpreter-dependent software ( see the note below ) |
|
1731 |
+
| Cuda | Toolkits, cudnn and related libraries |
|
1732 |
+
| Pytorch/Tensorflow | Pytorch/TF built with a specific Cuda/Cudnn version for Mila’s GPUs ( see the related paragraph ) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1733 |
|
1734 |
Note
|
1735 |
Modules which are nested (../../..) usually depend on other software/module
|
|
|
1990 |
apt-get update
|
1991 |
apt-get install -y cmake libcupti-dev libyaml-dev wget unzip
|
1992 |
apt-get clean
|
1993 |
+
echo ""Instal"
|
1994 |
+
Second way: Use recipes,https://docs.mila.quebec/Userguide.html#second-way-use-recipes,"ling things with pip""
|
1995 |
pip install tqdm
|
1996 |
echo ""Creating mount points""
|
1997 |
mkdir /dataset
|
|
|
2020 |
You always need to use sudo when you build a container from a
|
2021 |
recipe. As there is no access to sudo on the cluster, a personal computer or
|
2022 |
the use singularity hub is needed to build a container
|
|
|
2023 |
"
|
2024 |
Build recipe on singularity hub,https://docs.mila.quebec/Userguide.html#build-recipe-on-singularity-hub,"Build recipe on singularity hub
|
2025 |
Singularity hub allows users to build containers from recipes directly on
|
|
|
2095 |
mkdir /Gym && cd /Gym
|
2096 |
git clone https://github.com/openai/gym.git || true && \
|
2097 |
mkdir /Gym/.mujoco && cd /Gym/.mujoco
|
2098 |
+
wget https://www.roboti.us/do"
|
2099 |
+
"Example: Recipe with OpenAI gym, MuJoCo and Miniworld",https://docs.mila.quebec/Userguide.html#example-recipe-with-openai-gym-mujoco-and-miniworld,"wnload/mjpro150_linux.zip && \
|
2100 |
unzip mjpro150_linux.zip && \
|
2101 |
wget https://www.roboti.us/download/mujoco200_linux.zip && \
|
2102 |
unzip mujoco200_linux.zip && \
|
|
|
2106 |
export MUJOCO_PY_MJKEY_PATH=/Gym/.mujoco/mjkey.txt
|
2107 |
export MUJOCO_PY_MUJOCO_PATH=/Gym/.mujoco/mujoco150/
|
2108 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mjpro150/bin
|
2109 |
+
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mujoco200/bin
|
|
|
2110 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/bin
|
2111 |
cp /mjkey.txt /Gym/.mujoco/mjkey.txt
|
2112 |
# Install Python dependencies
|
|
|
2127 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mjpro150/bin
|
2128 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mujoco200/bin
|
2129 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/bin
|
2130 |
+
export PATH=/Gym/gym/.tox/py3/bin:$PATH"
|
2131 |
+
"Example: Recipe with OpenAI gym, MuJoCo and Miniworld",https://docs.mila.quebec/Userguide.html#example-recipe-with-openai-gym-mujoco-and-miniworld,"
|
2132 |
|
2133 |
%runscript
|
2134 |
exec /bin/sh ""$@""
|
|
|
2170 |
|
2171 |
# Download Gym and MuJoCo
|
2172 |
mkdir /Gym && cd /Gym
|
2173 |
+
git clone https://github.com/openai/gym.git || true && \
|
|
|
2174 |
mkdir /Gym/.mujoco && cd /Gym/.mujoco
|
2175 |
wget https://www.roboti.us/download/mjpro150_linux.zip && \
|
2176 |
unzip mjpro150_linux.zip && \
|
|
|
2180 |
|
2181 |
# Export global environment variables
|
2182 |
export MUJOCO_PY_MJKEY_PATH=/Gym/.mujoco/mjkey.txt
|
2183 |
+
export MUJOCO_PY_MUJOCO_PATH=/Gym/.mujoco/mujo"
|
2184 |
+
"Example: Recipe with OpenAI gym, MuJoCo and Miniworld",https://docs.mila.quebec/Userguide.html#example-recipe-with-openai-gym-mujoco-and-miniworld,"co150/
|
2185 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mjpro150/bin
|
2186 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/Gym/.mujoco/mujoco200/bin
|
2187 |
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/bin
|
|
|
2218 |
|
2219 |
Keep in mind that those environment variables are sourced at runtime and not at
|
2220 |
build time. This is why, you should also define them in the %post section
|
2221 |
+
since they are required to install MuJoCo"
|
|
|
2222 |
Using containers on clusters,https://docs.mila.quebec/Userguide.html#using-containers-on-clusters,"Using containers on clusters
|
2223 |
"
|
2224 |
How to use containers on clusters,https://docs.mila.quebec/Userguide.html#how-to-use-containers-on-clusters,"How to use containers on clusters
|
|
|
2663 |
server that will redirect you to a session as soon as it is available.
|
2664 |
It is currently available for Mila clusters and some Digital Research Alliance
|
2665 |
of Canada (Alliance) clusters.
|
2666 |
+
| Cluster | Address | Login type |
|
2667 |
+
|------------|---------------------------------------------|--------------|
|
2668 |
+
| Mila Local | https://jupyterhub.server.mila.quebec | Google Oauth |
|
2669 |
+
| Alliance | https://docs.alliancecan.ca/wiki/JupyterHub | DRAC login |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2670 |
|
2671 |
Warning
|
2672 |
Do not forget to close the JupyterLab session! Closing the window leaves
|
|
|
2827 |
|
2828 |
|
2829 |
--exclusive is important to specify subsequent step/srun to bind to different cpus.
|
2830 |
+
This will produce 8 output files"
|
2831 |
+
Sharing a node with multiple GPU & multiple processes/GPU,https://docs.mila.quebec/Userguide.html#sharing-a-node-with-multiple-gpu-multiple-processes-gpu,", 2 for each step:
|
2832 |
|
2833 |
JOBID-step-0-task-0.out
|
2834 |
JOBID-step-0-task-1.out
|
|
|
2849 |
0: | 0 Tesla P100-PCIE... On | 00000000:82:00.0 Off | 0 |
|
2850 |
1: | 0 Tesla P100-PCIE... On | 00000000:82:00.0 Off | 0 |
|
2851 |
0: | 0 Tesla P100-PCIE... On | 00000000:03:00.0 Off | 0 |
|
2852 |
+
1: | 0 Tesla P100-PCIE... On | 00000000:03:00.0 Off | 0 |"
|
|
|
2853 |
Multiple Nodes,https://docs.mila.quebec/Userguide.html#multiple-nodes,"Multiple Nodes
|
2854 |
"
|
2855 |
Data Parallel,https://docs.mila.quebec/Userguide.html#data-parallel,"Data Parallel
|
buster/docparser.py
CHANGED
@@ -2,6 +2,7 @@ import glob
|
|
2 |
import math
|
3 |
import os
|
4 |
|
|
|
5 |
import pandas as pd
|
6 |
import tiktoken
|
7 |
from bs4 import BeautifulSoup
|
@@ -14,7 +15,20 @@ EMBEDDING_ENCODING = "cl100k_base" # this the encoding for text-embedding-ada-0
|
|
14 |
BASE_URL = "https://docs.mila.quebec/"
|
15 |
|
16 |
|
17 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
"""Parse all HTML files in `root_dir`, and extract all sections.
|
19 |
|
20 |
Sections are broken into subsections if they are longer than `max_section_length`.
|
@@ -34,11 +48,10 @@ def get_all_documents(root_dir: str, max_section_length: int = 3000) -> pd.DataF
|
|
34 |
|
35 |
# If sections has subsections, keep only the part before the first subsection
|
36 |
if len(section_href) > 1:
|
37 |
-
section_siblings = section_soup.section.previous_siblings
|
38 |
-
section =
|
39 |
-
section = "".join(section[::-1])[1:]
|
40 |
else:
|
41 |
-
section = section_soup.
|
42 |
|
43 |
url = section_found["href"]
|
44 |
name = section_found.parent.text[:-1]
|
|
|
2 |
import math
|
3 |
import os
|
4 |
|
5 |
+
import bs4
|
6 |
import pandas as pd
|
7 |
import tiktoken
|
8 |
from bs4 import BeautifulSoup
|
|
|
15 |
BASE_URL = "https://docs.mila.quebec/"
|
16 |
|
17 |
|
18 |
+
def parse_section(nodes: list[bs4.element.NavigableString]) -> str:
|
19 |
+
section = []
|
20 |
+
for node in nodes:
|
21 |
+
if node.name == "table":
|
22 |
+
node_text = pd.read_html(node.prettify())[0].to_markdown(index=False, tablefmt="github")
|
23 |
+
else:
|
24 |
+
node_text = node.text
|
25 |
+
section.append(node_text)
|
26 |
+
section = "".join(section)[1:]
|
27 |
+
|
28 |
+
return section
|
29 |
+
|
30 |
+
|
31 |
+
def get_all_documents(root_dir: str, max_section_length: int = 2000) -> pd.DataFrame:
|
32 |
"""Parse all HTML files in `root_dir`, and extract all sections.
|
33 |
|
34 |
Sections are broken into subsections if they are longer than `max_section_length`.
|
|
|
48 |
|
49 |
# If sections has subsections, keep only the part before the first subsection
|
50 |
if len(section_href) > 1:
|
51 |
+
section_siblings = list(section_soup.section.previous_siblings)[::-1]
|
52 |
+
section = parse_section(section_siblings)
|
|
|
53 |
else:
|
54 |
+
section = parse_section(section_soup.children)
|
55 |
|
56 |
url = section_found["href"]
|
57 |
name = section_found.parent.text[:-1]
|