update dags

This commit is contained in:
2026-04-16 16:19:46 +03:00
parent dad7b2cd1a
commit 0a43170680

View File

@@ -40,7 +40,7 @@ def run_training_node_func(rank, world_size):
import time
# CRITICAL: Set NCCL environment variables BEFORE importing torch
os.environ['NCCL_DEBUG'] = 'INFO'
os.environ['NCCL_DEBUG'] = 'TRACE'
os.environ['NCCL_TIMEOUT'] = str(NCCL_TIMEOUT)
os.environ['TORCH_NCCL_BLOCKING_WAIT'] = '1'
os.environ['NCCL_IB_DISABLE'] = '0' # Enable InfiniBand