Skip to content

Enable megascale abort on hang or error #1010

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.

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

Merged
merged 2 commits into from
Feb 21, 2025
Merged
Changes from all commits
Commits
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
12 changes: 11 additions & 1 deletion axlearn/common/compiler_options.py
Original file line number Diff line number Diff line change
Expand Up @@ -126,6 +126,16 @@ def default_xla_options(
xla_tpu_data_parallel_opt_different_sized_ops="true",
# Group non-blocking DCN collectives into as few stages as possible.
xla_tpu_enable_sunk_dcn_allreduce_done_with_host_reduction="true",
# Aborting the coordinator after collecting errors from all workers.
# All workers will also abort after they detect the coordinator is shutdown.
megascale_error_reporter_abort_on_hang="true",
# Similar to megascale_error_reporter_abort_on_hang but for unrecoverable errors.
megascale_error_reporter_abort_on_error="true",
# Increase the timeout at which a hang is detected/reported, default is 5m.
megascale_graph_hang_threshold="10m",
# Similar to megascale_graph_hang_threshold but specific to within a launch_id.
# Default is 1m.
megascale_graph_within_launch_hang_threshold="10m",
)

# Validate options. Will never fail if this function is implemented correctly.
Expand All @@ -134,7 +144,7 @@ def default_xla_options(
int(v)
continue
except ValueError:
assert v in [True, False, "true", "false", "megachip_tccontrol"], (k, v)
assert v in [True, False, "true", "false", "megachip_tccontrol", "10m"], (k, v)

return options

Expand Down
Loading