Skip to content

Commit

Permalink
For undecimal process group name, replace raising exeption with warni…
Browse files Browse the repository at this point in the history
…ng message and continue.
  • Loading branch information
GSSBMW authored and TaekyungHeo committed Aug 5, 2024
1 parent aa7f260 commit 7dcd0eb
Showing 1 changed file with 5 additions and 1 deletion.
6 changes: 5 additions & 1 deletion et_replay/comm/commsTraceParser.py
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,9 @@
from et_replay.comm.backend.base_backend import supportedP2pOps
from et_replay.comm.comms_utils import commsArgs

import logging
logger = logging.getLogger(__name__)

tensorDtypeMap = {
"Tensor(int)": "int",
"Tensor(float)": "float",
Expand Down Expand Up @@ -112,7 +115,8 @@ def _parse_proc_group_info(in_trace: ExecutionTrace):
for pg in pg_objs:
if not pg["pg_name"].isdecimal():
# TODO support local synchronization pg
raise ValueError(f"Process group name is {pg['pg_name']} in node {node['id']}, which is not supported.")
logger.warning(f"Process group name is {pg['pg_name']} in node {node.id}, which is not supported. Skip.")
continue
(pg_id, ranks, group_size, group_count) = [pg[k] for k in ["pg_name", "ranks", "group_size", "group_count"]]
pg_id = int(pg_id)
pg_ranks_map[node.id][pg_id] = (
Expand Down

0 comments on commit 7dcd0eb

Please sign in to comment.