Skip to content

feat: add rank0 async HF save logs #1845

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
caixianzhang wants to merge 5 commits into
base: main
Choose a base branch
from
+12 −5
Open

feat: add rank0 async HF save logs #1845

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
d8b5a11
feat: add rank0 async HF logs
caixianzhang May 27, 2026
93b5ffe
fix: import time for async HF compose logs
caixianzhang May 27, 2026
98ad563
refactor: simplify async HF writer log path
caixianzhang May 27, 2026
182d039
refactor: rely on log timestamps for async HF writer
caixianzhang May 27, 2026
8bae23b
refactor: use log_rank0 for async HF writer logs
caixianzhang May 27, 2026
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
11 changes: 6 additions & 5 deletions xtuner/v1/engine/train_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -318,11 +318,12 @@ def async_save_hf(
save_dtype: torch.dtype = torch.bfloat16,
cleanup_hf_dirs: Sequence[str | Path] = (),
) -> AsyncHFSaveHandle:
return self.model.async_save_hf(
hf_dir=hf_dir,
save_dtype=save_dtype,
cleanup_hf_dirs=cleanup_hf_dirs,
)
with profile_time_and_memory(f"[Async saving HF to {hf_dir} launch cost]"):
return self.model.async_save_hf(
hf_dir=hf_dir,
save_dtype=save_dtype,
cleanup_hf_dirs=cleanup_hf_dirs,
)

def wait_async_hf(self, handle: AsyncHFSaveHandle | None = None) -> Path | None:
return self.model.wait_async_hf(handle)
Expand Down
3 changes: 3 additions & 0 deletions xtuner/v1/model/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -762,6 +762,7 @@ def _run_async_hf_writer(
cleanup_done_path: Path,
rank: int,
) -> None:
log_rank0.info(f"[Async saving HF to {tmp_hf_dir} writer] started")
try:
set_async_save_process_qos()
self._cleanup_async_hf_dirs_before_write(
Expand All @@ -775,7 +776,9 @@ def _run_async_hf_writer(
weight_map=weight_map,
status_path=status_path,
)
log_rank0.info(f"[Async saving HF to {tmp_hf_dir} writer] finished")
except Exception as exc:
log_rank0.error(f"[Async saving HF to {tmp_hf_dir} writer] failed: {exc}")
status = {"rank": rank, "ok": False, "error": str(exc), "weight_map": {}}
with status_path.open("w") as f:
f.write(json.dumps(status, indent=2))
Expand Down
3 changes: 3 additions & 0 deletions xtuner/v1/model/compose/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -258,6 +258,7 @@ def _run_async_hf_compose_writer(
cleanup_done_path: Path,
rank: int,
) -> None:
log_rank0.info(f"[Async saving HF to {tmp_hf_dir} writer] started")
try:
set_async_save_process_qos()
self._cleanup_async_hf_dirs_before_write(
Expand All @@ -270,7 +271,9 @@ def _run_async_hf_compose_writer(
status = {"rank": rank, "ok": True, "error": "", "weight_map": merged_weight_map}
with status_path.open("w") as f:
f.write(json.dumps(status, indent=2))
log_rank0.info(f"[Async saving HF to {tmp_hf_dir} writer] finished")
except Exception as exc:
log_rank0.error(f"[Async saving HF to {tmp_hf_dir} writer] failed: {exc}")
status = {"rank": rank, "ok": False, "error": str(exc), "weight_map": {}}
with status_path.open("w") as f:
f.write(json.dumps(status, indent=2))
Expand Down
Loading