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

Jump to bottom

Fix wrong inputs/model placement when using a single core #725

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix wrong inputs/model placement when using a single core #725

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
9 changes: 7 additions & 2 deletions optimum/neuron/modeling_traced.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,7 @@
NEURON_COMPILER_VERSION = get_neuroncc_version()

if is_neuronx_available():
import torch_neuronx
from torch_neuronx import move_trace_to_device

NEURON_COMPILER_TYPE = "neuronx-cc"
Expand Down Expand Up @@ -127,8 +128,12 @@ def load_model(
if path.is_file():
model = torch.jit.load(path)
# For non-inlined models, send the module manually to device. This is important for weights/neff non-inlined module since when loading the module, the neff is automatically moved to Neuron but not the weights. We need to move the weights to Neuron as well manually to avoid great host to device IO penalty.
if is_neuronx_available() and to_neuron:
move_trace_to_device(model, device_id)
if is_neuronx_available():
torch_neuronx.experimental.set_neuron_cores(
model, start_nc=0, nc_count=1
) # The inputs are allocated to nc:0 by default, this line ensures both input tensors and the model are on the same core.
if to_neuron:
move_trace_to_device(model, device_id)
return model

def replace_weights(self, weights: Optional[Union[Dict[str, torch.Tensor], torch.nn.Module]] = None):
Expand Down
Loading