George Stykalin 48abe40c15 Fix DDP master address for dynamic task scheduling ...

The fundamental issue was that MASTER_ADDR was hardcoded to
airflow-worker-gpu-0, but rank 0 might run on any worker due to
Celery's dynamic task scheduling.

Now rank 0 dynamically stores its FQDN as the master address,
and all other ranks connect to it.

Co-Authored-By: Claude <noreply@anthropic.com>

2026-04-15 12:23:15 +03:00

..

connection-check-default.py

update graphs

2026-01-26 16:13:01 +03:00

connection-check.py

update dags

2026-01-26 15:49:45 +03:00

test-train-pytorch.py

Fix DDP master address for dynamic task scheduling

2026-04-15 12:23:15 +03:00

test-train-tensorflow.py

init commit

2026-01-12 16:25:19 +03:00

train.py

update dags

2026-04-15 12:10:19 +03:00