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

Add support for PyTorch GPU memory limit #41

Merged
merged 11 commits into from
Jan 18, 2023
Merged

Add support for PyTorch GPU memory limit #41

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
0dd0d43
Add support for PyTorch GPU memory limit
kcm Jan 3, 2023
1072d9e
log bad CUDA memory setting as an error
kcm Jan 5, 2023
c599680
Add support for sentence-t5-base model
antas-marcin Jan 13, 2023
412eb3f
Switch to github actions
antas-marcin Jan 13, 2023
d31f0c1
Fixed deploy step
antas-marcin Jan 14, 2023
36e4e04
Fix GIT_TAG env variable in deploy step
antas-marcin Jan 14, 2023
e0514a6
Enable all of the transformers models
antas-marcin Jan 14, 2023
3d1d4c5
Add duplicate /vectors route to remove 307 redirects
trengrj Jan 16, 2023
4469e92
remove un-used packages, remove installation of third party libs (#38)
StefanBogdan Jan 16, 2023
e7f5788
more explicit variable naming for CUDA memory percentage
kcm Jan 17, 2023
43d0946
Merge branch 'main' into GH-39
kcm Jan 17, 2023
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: 9 additions & 1 deletion app.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,14 @@ def startup_event():
global meta_config

cuda_env = os.getenv("ENABLE_CUDA")
cuda_per_process_memory_fraction = 1.0
if "CUDA_PER_PROCESS_MEMORY_FRACTION" in os.environ:
try:
cuda_per_process_memory_fraction = float(os.getenv("CUDA_PER_PROCESS_MEMORY_FRACTION"))
except ValueError:
logger.error(f"Invalid CUDA_PER_PROCESS_MEMORY_FRACTION (should be between 0.0-1.0)")
if 0.0 <= cuda_per_process_memory_fraction <= 1.0:
logger.info(f"CUDA_PER_PROCESS_MEMORY_FRACTION set to {cuda_per_process_memory_fraction}")
cuda_support=False
cuda_core=""

Expand All @@ -30,7 +38,7 @@ def startup_event():
logger.info("Running on CPU")

meta_config = Meta('./models/model')
vec = Vectorizer('./models/model', cuda_support, cuda_core,
vec = Vectorizer('./models/model', cuda_support, cuda_core, cuda_per_process_memory_fraction,
meta_config.getModelType(), meta_config.get_architecture())


Expand Down
5 changes: 4 additions & 1 deletion vectorizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,16 +33,19 @@ class Vectorizer:
cuda_core: str
model_type: str

def __init__(self, model_path: str, cuda_support: bool, cuda_core: str, model_type: str, architecture: str):
def __init__(self, model_path: str, cuda_support: bool, cuda_core: str, cuda_per_process_memory_fraction: float, model_type: str, architecture: str):
self.cuda = cuda_support
self.cuda_core = cuda_core
self.cuda_per_process_memory_fraction = cuda_per_process_memory_fraction
self.model_type = model_type

self.model_delegate: HFModel = ModelFactory.model(model_type, architecture)
self.model = self.model_delegate.create_model(model_path)

if self.cuda:
self.model.to(self.cuda_core)
if self.cuda_per_process_memory_fraction:
torch.cuda.set_per_process_memory_fraction(self.cuda_per_process_memory_fraction)
self.model.eval() # make sure we're in inference mode, not training

self.tokenizer = self.model_delegate.create_tokenizer(model_path)
Expand Down