chore: add use_gpu for cifar finetuning #882

kcelia · 2024-09-23T13:14:24Z

No description provided.

andrei-stoian-zama

Thanks, but you need to use the new CML GPU api instead of the Concrete one

use_case_examples/cifar/cifar_brevitas_finetuning/cifar_utils.py

use_case_examples/cifar/cifar_brevitas_finetuning/PerrorImpactOnFMNIST.ipynb

use_case_examples/resnet/README.md

andrei-stoian-zama

@kcelia Can you please confirm you can run these notebooks on your machine ?

use_case_examples/cifar/cifar_brevitas_finetuning/FromImageNetToCifar.ipynb

use_case_examples/cifar/cifar_brevitas_finetuning/CifarQuantizationAwareTraining.ipynb

use_case_examples/cifar/cifar_brevitas_finetuning/CifarInFheWithSmallerAccumulators.ipynb

use_case_examples/cifar/cifar_brevitas_finetuning/CifarInFhe.ipynb

jfrery · 2024-09-24T15:02:18Z

use_case_examples/resnet/run_resnet18_fhe.py

-    model, images, n_bits, rounding_threshold_bits=None, fhe_mode="disable", use_gpu=False
+    model,
+    images,
+    n_bits,
+    rounding_threshold_bits=None,
+    fhe_mode="disable",
+    compilation_device="cpu",


I preferred use_gpu or just device=... to do the same as torch.

"@jfrery, use_device might sound like a boolean type, but in the compilation functions, it's actually a string that can be either 'cpu' or 'gpu'.

The reason it's not simply named device is because the PyTorch device and the CML one may differ. This distinction helps differentiate between the device used for PyTorch training/evaluation and the one used for CML compilation.

@jfrery the API was decided some time ago:

https://zama-ai.slack.com/archives/C0455QDGT6C/p1723040023589979

https://zama-ai.slack.com/archives/C0455QDGT6C/p1724860575912049

device="cuda|cpu" is perfect yes but here we have compilation_device.

The reason it's not simply named device is because the PyTorch device and the CML one may differ.

Well as far as I know we try to stay close to the torch API. We did it for sklearn and the others. I don't see why now we would want to differ. device seems the right choice here.

We assume that the compile_device may differ from the device used to train the model.

Additionally, on standard machines, setting compile_device="cpu" is currently faster than compile_device="cuda". So, it's better to keep them separate.

andrei-stoian-zama

one small error remains

use_case_examples/cifar/cifar_brevitas_training/evaluate_torch_cml.py

use_case_examples/cifar/cifar_brevitas_finetuning/CifarQuantizationAwareTraining.ipynb

kcelia · 2024-09-25T12:30:39Z

I tested on my machine, all the scripts are working

github-actions · 2024-09-25T13:50:14Z

⚠️ Known flaky tests have been rerun ⚠️

One or several tests initially failed but were identified as known flaky. tests. Therefore, they have been rerun and passed. See below for more details.

Failed tests details

Known flaky tests that initially failed:

tests/torch/test_compile_torch.py::test_compile_torch_or_onnx_conv_networks[True-True-CNN-relu]- tests/torch/test_compile_torch.py::test_compile_torch_or_onnx_conv_networks[False-True-CNN-relu]- tests/torch/test_compile_torch.py::test_compile_torch_or_onnx_conv_networks[False-True-CNN_grouped-relu]

github-actions · 2024-09-25T13:50:16Z

Coverage failed ❌

Coverage details

---------- coverage: platform linux, python 3.8.18-final-0 -----------
Name                                 Stmts   Miss  Cover   Missing
------------------------------------------------------------------
src/concrete/ml/onnx/onnx_utils.py      59     10    83%   594-620
------------------------------------------------------------------
TOTAL                                 8138     10    99%

59 files skipped due to complete coverage.

chore: add use_gpu for cifar finetuning

9b1c77f

kcelia self-assigned this Sep 23, 2024

cla-bot bot added the cla-signed label Sep 23, 2024

andrei-stoian-zama requested changes Sep 23, 2024

View reviewed changes

use_case_examples/cifar/cifar_brevitas_finetuning/cifar_utils.py Outdated Show resolved Hide resolved

use_case_examples/cifar/cifar_brevitas_finetuning/PerrorImpactOnFMNIST.ipynb Outdated Show resolved Hide resolved

use_case_examples/resnet/README.md Show resolved Hide resolved

kcelia force-pushed the gpu_update_4608 branch from 094545e to e57809c Compare September 23, 2024 14:21

andrei-stoian-zama requested changes Sep 24, 2024

View reviewed changes

kcelia force-pushed the gpu_update_4608 branch 2 times, most recently from fc37f69 to 7fb522d Compare September 24, 2024 14:14

kcelia added 3 commits September 24, 2024 16:17

chore: update cifar finetuning

17bdca2

chore: add use_gpu to cifar training

b5e215d

chore: add use_gpu to resnet18

46b831e

kcelia force-pushed the gpu_update_4608 branch from 7fb522d to 46b831e Compare September 24, 2024 14:18

jfrery reviewed Sep 24, 2024

View reviewed changes

kcelia force-pushed the gpu_update_4608 branch from a7b37a4 to 218c666 Compare September 25, 2024 07:35

andrei-stoian-zama requested changes Sep 25, 2024

View reviewed changes

use_case_examples/cifar/cifar_brevitas_training/evaluate_torch_cml.py Outdated Show resolved Hide resolved

use_case_examples/cifar/cifar_brevitas_finetuning/CifarQuantizationAwareTraining.ipynb Outdated Show resolved Hide resolved

kcelia force-pushed the gpu_update_4608 branch from 40bc329 to 1bf1cad Compare September 25, 2024 12:30

kcelia requested a review from andrei-stoian-zama September 25, 2024 13:57

kcelia marked this pull request as ready for review September 25, 2024 13:57

kcelia requested a review from a team as a code owner September 25, 2024 13:57

kcelia force-pushed the gpu_update_4608 branch from 1bf1cad to f4f7c30 Compare September 25, 2024 14:51

chore: update

e4934c8

kcelia force-pushed the gpu_update_4608 branch from f4f7c30 to e4934c8 Compare September 25, 2024 14:52

andrei-stoian-zama approved these changes Sep 25, 2024

View reviewed changes

kcelia merged commit b37ceed into main Sep 25, 2024
17 of 18 checks passed

kcelia deleted the gpu_update_4608 branch September 25, 2024 16:43

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

chore: add use_gpu for cifar finetuning #882

chore: add use_gpu for cifar finetuning #882

kcelia commented Sep 23, 2024

andrei-stoian-zama left a comment

andrei-stoian-zama left a comment

jfrery Sep 24, 2024

kcelia Sep 25, 2024

andrei-stoian-zama Sep 25, 2024

jfrery Sep 25, 2024

kcelia Sep 25, 2024

andrei-stoian-zama left a comment

kcelia commented Sep 25, 2024 •

edited

Loading

github-actions bot commented Sep 25, 2024

Known flaky tests that initially failed:

github-actions bot commented Sep 25, 2024

chore: add use_gpu for cifar finetuning #882

chore: add use_gpu for cifar finetuning #882

Conversation

kcelia commented Sep 23, 2024

andrei-stoian-zama left a comment

Choose a reason for hiding this comment

andrei-stoian-zama left a comment

Choose a reason for hiding this comment

jfrery Sep 24, 2024

Choose a reason for hiding this comment

kcelia Sep 25, 2024

Choose a reason for hiding this comment

andrei-stoian-zama Sep 25, 2024

Choose a reason for hiding this comment

jfrery Sep 25, 2024

Choose a reason for hiding this comment

kcelia Sep 25, 2024

Choose a reason for hiding this comment

andrei-stoian-zama left a comment

Choose a reason for hiding this comment

kcelia commented Sep 25, 2024 • edited Loading

github-actions bot commented Sep 25, 2024

⚠️ Known flaky tests have been rerun ⚠️

Known flaky tests that initially failed:

github-actions bot commented Sep 25, 2024

Coverage failed ❌

kcelia commented Sep 25, 2024 •

edited

Loading