Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
B
Berzelius Benchmarks
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Requirements
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Locked files
Build
Pipelines
Jobs
Pipeline schedules
Test cases
Artifacts
Deploy
Releases
Package registry
Model registry
Operate
Environments
Terraform modules
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Code review analytics
Issue analytics
Insights
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Terms and privacy
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
Xuan Gu
Berzelius Benchmarks
Commits
13aabe3e
Commit
13aabe3e
authored
1 year ago
by
Xuan Gu
Browse files
Options
Downloads
Patches
Plain Diff
Update 2 files
- /scripts/benchmark_sbatch_submit.sh - /README.md
parent
d01afc55
No related branches found
No related tags found
No related merge requests found
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
README.md
+7
-2
7 additions, 2 deletions
README.md
scripts/benchmark_sbatch_submit.sh
+8
-4
8 additions, 4 deletions
scripts/benchmark_sbatch_submit.sh
with
15 additions
and
6 deletions
README.md
+
7
−
2
View file @
13aabe3e
...
@@ -83,9 +83,14 @@ bash scripts/copy_data_for_benchmark.sh
...
@@ -83,9 +83,14 @@ bash scripts/copy_data_for_benchmark.sh
You can choose either singularity or enroot in the scripts
```benchmark_single_node.sbatch```
and
```benchmark_multi_node.sbatch```
.
You can choose either singularity or enroot in the scripts
```benchmark_single_node.sbatch```
and
```benchmark_multi_node.sbatch```
.
The input argument is the number of nodes used for benchmarking.
The input arguments are:
1.
Number of nodes,
2.
Number of gpus per node,
3.
Number of iterations for each parameter setting.
We will average the benchmark performance over the iterations. We use a batch size of 128 which is the maximum usable batch size without a OOM error.
```
```
bash scripts/benchmark_sbatch_submit.sh 1
bash scripts/benchmark_sbatch_submit.sh 1
8 100
```
```
### Results
### Results
...
...
This diff is collapsed.
Click to expand it.
scripts/benchmark_sbatch_submit.sh
+
8
−
4
View file @
13aabe3e
#!/bin/bash
mkdir
-p
sbatch_out
mkdir
-p
sbatch_out
dim
=
2
dim
=
2
NUM_NODES
=
$1
NUM_NODES
=
$1
NUM_GPUS
=
$2
NUM_ITERATIONS
=
$3
if
[
$NUM_NODES
-eq
1
]
;
then
if
[
$NUM_NODES
-eq
1
]
;
then
for
nodes
in
{
1..1
}
;
do
for
nodes
in
{
1..1
}
;
do
for
gpus
in
{
1..
8
}
;
do
for
gpus
in
{
1..
$NUM_GPUS
}
;
do
for
batch_size
in
128
;
do
for
batch_size
in
128
;
do
for
iteration
in
{
1..
100
}
;
do
for
iteration
in
{
1..
$NUM_ITERATIONS
}
;
do
echo
dim
${
dim
}
, nodes
${
nodes
}
, gpus
${
gpus
}
, batch_size
${
batch_size
}
, iteration
${
iteration
}
echo
dim
${
dim
}
, nodes
${
nodes
}
, gpus
${
gpus
}
, batch_size
${
batch_size
}
, iteration
${
iteration
}
# For single node
# For single node
...
@@ -20,9 +24,9 @@ if [ $NUM_NODES -eq 1 ]; then
...
@@ -20,9 +24,9 @@ if [ $NUM_NODES -eq 1 ]; then
done
done
else
else
for
nodes
in
{
2..
$NUM_NODES
}
;
do
for
nodes
in
{
2..
$NUM_NODES
}
;
do
for
gpus
in
{
8
}
;
do
for
gpus
in
{
$NUM_GPUS
}
;
do
for
batch_size
in
128
;
do
for
batch_size
in
128
;
do
for
iteration
in
{
1..
100
}
;
do
for
iteration
in
{
1..
$NUM_ITERATIONS
}
;
do
# For multi node
# For multi node
sbatch
-o
sbatch_out/benchmark_dim
${
dim
}
_nodes
${
nodes
}
_gpus
${
gpus
}
_batchsize
${
batch_size
}
_iteration
${
iteration
}
.out scripts/benchmark_multi_node.sbatch
${
dim
}
${
nodes
}
${
gpus
}
${
batch_size
}
${
iteration
}
sbatch
-o
sbatch_out/benchmark_dim
${
dim
}
_nodes
${
nodes
}
_gpus
${
gpus
}
_batchsize
${
batch_size
}
_iteration
${
iteration
}
.out scripts/benchmark_multi_node.sbatch
${
dim
}
${
nodes
}
${
gpus
}
${
batch_size
}
${
iteration
}
...
...
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment