Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Multi node tweaks #248

Merged
merged 7 commits into from
Aug 15, 2024
Merged

Multi node tweaks #248

merged 7 commits into from
Aug 15, 2024

Conversation

Delaunay
Copy link
Collaborator

@Delaunay Delaunay commented Aug 5, 2024

No description provided.

@Delaunay
Copy link
Collaborator Author

Delaunay commented Aug 7, 2024

=================
Benchmark results
=================
bench                          | fail |   n |       perf |   sem% |   std% | peak_memory |      score | weight
brax                           |    0 |   1 |  705221.22 |   0.2% |   1.4% |        2670 |  705221.22 |   1.00
dinov2-giant-gpus              |    0 |   1 |     290.27 |   1.4% |  11.0% |       69546 |     290.27 |   1.00
dinov2-giant-nodes             |    1 |   2 |        nan |   nan% |   nan% |         nan |        nan |   0.00
dinov2-giant-single            |    0 |   8 |      45.93 |   0.9% |  19.5% |       74646 |     371.12 |   0.00
bf16                           |    0 |   8 |     293.43 |   0.2% |   6.2% |        1788 |    2360.98 |   0.00
fp16                           |    0 |   8 |     290.16 |   0.1% |   3.8% |        1788 |    2329.58 |   0.00
fp32                           |    0 |   8 |      19.13 |   0.0% |   0.7% |        2166 |     153.10 |   0.00
tf32                           |    0 |   8 |     146.89 |   0.1% |   3.7% |        2166 |    1179.13 |   0.00
bert-fp16                      |    0 |   8 |     260.00 |   1.1% |  17.4% |         nan |    2134.01 |   0.00
bert-fp32                      |    0 |   8 |      44.57 |   0.7% |  10.4% |       21170 |     362.77 |   0.00
bert-tf32                      |    0 |   8 |     140.28 |   1.0% |  15.0% |        1704 |    1148.08 |   0.00
bert-tf32-fp16                 |    0 |   8 |     259.09 |   1.2% |  18.3% |         nan |    2125.99 |   3.00
reformer                       |    0 |   8 |      61.96 |   0.3% |   6.4% |       25404 |     499.36 |   1.00
t5                             |    0 |   8 |      50.01 |   0.6% |  12.9% |       34390 |     404.81 |   2.00
whisper                        |    0 |   8 |     425.67 |   1.1% |  24.5% |         nan |    3438.69 |   1.00
lightning                      |    0 |   8 |     467.95 |   1.5% |  33.5% |       27360 |    3772.13 |   0.00
lightning-gpus                 |    0 |   1 |    1937.94 |   4.4% |  34.8% |       28178 |    1937.94 |   1.00
llama                          |    0 |   8 |     429.17 |   4.5% |  81.4% |       27820 |    3276.65 |   1.00
llm-full-mp-gpus               |    0 |   1 |      86.50 |   2.6% |  13.7% |       48400 |      86.50 |   1.00
llm-full-mp-nodes              |    1 |   2 |      49.97 |   1.6% |   8.4% |       35658 |      24.99 |   1.00
llm-lora-ddp-gpus              |    0 |   1 |    8995.26 |   1.2% |   6.2% |       36988 |    8995.26 |   1.00
llm-lora-ddp-nodes             |    1 |   2 |    4493.43 |   0.5% |   2.3% |       35678 |    2246.71 |   1.00
llm-lora-mp-gpus               |    0 |   1 |    1146.41 |   1.5% |   7.7% |       55972 |    1146.41 |   1.00
llm-lora-single                |    0 |   8 |    2717.90 |   0.2% |   3.1% |       49926 |   21797.19 |   1.00
super-slomo                    |    0 |   8 |      44.44 |   1.2% |  26.1% |       65928 |     358.16 |   1.00
focalnet                       |    0 |   8 |     341.79 |   0.9% |  20.2% |       23536 |    2765.88 |   2.00
convnext_large-fp16            |    0 |   8 |     331.04 |   1.2% |  18.6% |       27376 |    2714.04 |   0.00
convnext_large-fp32            |    0 |   8 |      59.16 |   0.7% |  11.2% |       55950 |     482.25 |   0.00
convnext_large-tf32            |    0 |   8 |     154.15 |   1.0% |  14.9% |       49650 |    1263.82 |   0.00
convnext_large-tf32-fp16       |    0 |   8 |     331.46 |   1.2% |  18.5% |       27376 |    2717.71 |   3.00
regnet_y_128gf                 |    0 |   8 |     115.71 |   0.6% |  12.4% |       29762 |     936.62 |   2.00
resnet152-ddp-gpus             |    0 |   1 |    2040.41 |   5.1% |  38.9% |       27988 |    2040.41 |   0.00
resnet50                       |    0 |   8 |     782.16 |   1.6% |  34.6% |       14848 |    6298.85 |   1.00
resnet50-noio                  |    0 |   8 |    1161.02 |   0.3% |   6.8% |       27480 |    9363.26 |   0.00

Scores
------
Failure rate:       1.46% (FAIL)
Score:            1711.32

Errors
------
3 errors, details in HTML report.

@Delaunay Delaunay changed the base branch from master to staging August 13, 2024 13:14
@Delaunay Delaunay marked this pull request as ready for review August 15, 2024 14:57
@Delaunay Delaunay merged commit 334e2b3 into mila-iqia:staging Aug 15, 2024
@Delaunay Delaunay deleted the rlhf branch August 15, 2024 14:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant