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.

Sign up for GitHub

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

[optim] Exclude timm_vision_transformer pt2, fix runtime errors in 1888 #1890

Closed
wants to merge 1 commit into from
Closed

[optim] Exclude timm_vision_transformer pt2, fix runtime errors in 1888 #1890

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 7 additions & 3 deletions userbenchmark/optim/run.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,7 @@ def get_unstable_models() -> Set[str]:
# PT2 dynamo tracing for the for-loop implementation takes over 30s.
# This is known + NOT going to be improved anytime soon, see
# https://github.com/pytorch/torchdynamo/issues/1803#issuecomment-1336688894
MODELS_TO_RUN_ON_PT2: List[str] = ['resnet18', 'timm_vision_transformer', 'timm_vision_transformer_large']
MODELS_TO_RUN_ON_PT2: List[str] = ['resnet18', 'timm_vision_transformer_large']

# NOTE: While it is possible to run these benchmarks on CPU, we skip running on CPU in CI because CPU stats can be
# unstable and we had stopped reporting them. You'll still be able to use this script to run CPU though, as it may
Expand Down Expand Up @@ -213,6 +213,7 @@ def get_unstable_models() -> Set[str]:
'sage',
'sam',
'shufflenet_v2_x1_0',
'simple_gpt',
'soft_actor_critic',
'speech_transformer',
'squeezenet1_1',
Expand Down Expand Up @@ -250,7 +251,10 @@ def get_unstable_models() -> Set[str]:
# 16h currently OOMs, but once it supports train, we should remove this line
# See tracker https://github.com/pytorch/benchmark/issues/1793
{'model': 'llama_v2_7b_16h'}
] +[
] + [
# Model needs to be run via dynamo torchbench and be provided distributed parameters
{'model': 'simple_gpt'}
] + [
# SparseAdam does not support dense gradients
{'optim': 'SparseAdam', 'model': m} for m in DENSE_MODELS
] + [
Expand Down Expand Up @@ -306,7 +310,7 @@ def get_unstable_models() -> Set[str]:
# See GH issue: https://github.com/pytorch/pytorch/issues/97361
{'model': m, 'device': 'cuda', 'func_str': 'pt2_', 'defaults': [df], 'optim': 'NAdam'} for m in [
'densenet121', 'doctr_reco_predictor', 'fambench_xlmr', 'hf_Bart', 'hf_Bert_large', 'hf_GPT2_large','hf_Longformer',
'hf_T5_base', 'hf_T5_large', 'moco', 'resnet152', 'timm_vision_transformer', 'yolov3'
'hf_T5_base', 'hf_T5_large', 'moco', 'resnet152', 'timm_vision_transformer', 'timm_vision_transformer_large', 'yolov3'
] for df in ['no_foreach', 'differentiable']
] + [
# torch.compile()'d optimizer.step() has too many arguments in the generated
Expand Down
Loading