Skip to content

[NPU] Fix bugs on NPU #385

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

Closed
lowdy1 wants to merge 1 commit into from
Closed

[NPU] Fix bugs on NPU #385

Show file tree
Hide file tree
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
3 changes: 2 additions & 1 deletion roll/third_party/vllm/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,7 +59,8 @@ async def create_async_llm(resource_placement_groups: List[Dict], **kwargs):
os.environ["PYTORCH_CUDA_ALLOC_CONF"] = ""
# torch.cuda may already init, explicitly disable expandable_segments
# here (only matters when VLLM_USE_RAY_SPMD_WORKER=0)
current_platform.memory._set_allocator_settings("expandable_segments:False")
if not current_platform.is_npu():
current_platform.memory._set_allocator_settings("expandable_segments:False")

os.environ["VLLM_CACHE_ROOT"] = os.path.join(get_default_cache_root(), "vllm", os.environ.get("WORKER_NAME", ""))

Expand Down
2 changes: 1 addition & 1 deletion roll/third_party/vllm/worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,7 +118,7 @@ def broadcast_parameter(self, names, dtypes, shapes, group_name, is_lora=False):
weights_and_handles = []
for name, dtype, shape in zip(names, dtypes, shapes):
target_dtype = dtype if isinstance(dtype, torch.dtype) else getattr(torch, dtype)
weight = torch.empty(shape, dtype=target_dtype, device=self.device)
weight = torch.empty(shape, dtype=target_dtype, device=current_platform.device_type)
handle = collective.broadcast(tensor=weight, src_rank=0, group_name=group_name, async_op=True)
weights_and_handles.append((name, weight, handle))

Expand Down
Loading