Skip to content

feat: make use_inplace_pin_memory configurable via env and request #80

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

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

feat: make use_inplace_pin_memory configurable via env and request #80

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
7 changes: 6 additions & 1 deletion checkpoint_engine/api.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,7 @@ def _init_api(ps: ParameterServer) -> Any:

class RegisterRequest(BaseModel):
files: list[str]
use_inplace_pin_memory: bool | None = None

class UpdateRequest(BaseModel):
ranks: list[int] = []
Expand All @@ -65,7 +66,11 @@ def wrap_exception(func: Callable[[], None]) -> Response:

@app.post("/v1/checkpoints/{checkpoint_name}/files")
async def register_files(checkpoint_name: str, req: RegisterRequest, raw: Request) -> Response:
return wrap_exception(lambda: ps.register_checkpoint(checkpoint_name, files=req.files))
return wrap_exception(
lambda: ps.register_checkpoint(
checkpoint_name, files=req.files, use_inplace_pin_memory=req.use_inplace_pin_memory
)
)

@app.delete("/v1/checkpoints/{checkpoint_name}")
async def unregister_checkpoint(checkpoint_name: str) -> Response:
Expand Down
8 changes: 7 additions & 1 deletion checkpoint_engine/ps.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -263,7 +263,7 @@ def register_checkpoint(
files: list[str] | None = None,
named_tensors: dict[str, torch.Tensor] | None = None,
use_shared_memory_pool: bool = False,
use_inplace_pin_memory: bool = True,
use_inplace_pin_memory: bool | None = None,
) -> None:
"""
Register a checkpoint to the parameter server. Both files and named_tensors will be registered together.
Expand All @@ -282,6 +282,12 @@ def register_checkpoint(
use_inplace_pin_memory: If True (default), allows inplace pin memory for /dev/shm/ safetensors files.
This option is ignored when ``use_shared_memory_pool`` is True.
"""
if use_inplace_pin_memory is None:
env_str = os.getenv("PS_USE_INPLACE_PIN_MEMORY", "true")
use_inplace_pin_memory = env_str.lower() in ["true", "1", "yes", "y"]
logger.info(
f"[rank{self._rank}] use_inplace_pin_memory set to {use_inplace_pin_memory} by environment variable PS_USE_INPLACE_PIN_MEMORY={env_str}"
)
if self.device_manager.device_type != "cuda" and use_inplace_pin_memory:
logger.warning(
f"[rank{self._rank}] Only cuda devices support in-place pin memory, set use_inplace_pin_memory to False"
Expand Down