From c61098cd3b19cd8bf872b1922d20333e7b63f454 Mon Sep 17 00:00:00 2001 From: speriaswamy-amd Date: Fri, 14 Aug 2026 16:11:29 -0400 Subject: [PATCH] fix(megatron_training_lib): compare verify_network_errors as the string it is self.verify_network_errors is always the string 'True'/'False' parsed from JSON, so `is True` can never be true and the post-run RDMA/ethtool error-counter comparison silently never runs. Matches the working `== 'True'` pattern already used for the same field elsewhere. AIMVT-276. --- cvs/lib/megatron_training_lib.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/cvs/lib/megatron_training_lib.py b/cvs/lib/megatron_training_lib.py index 03dff9631..f87ae8d18 100644 --- a/cvs/lib/megatron_training_lib.py +++ b/cvs/lib/megatron_training_lib.py @@ -781,7 +781,7 @@ def verify_training_results( # Check if RDMA and Ethtool stats have errors .. if self.distributed_training is True: - if self.verify_network_errors is True: + if self.verify_network_errors == 'True': self.rdma_stats_dict_after = linux_utils.get_rdma_stats_dict(self.phdl) self.ethtool_stats_dict_after = linux_utils.get_nic_ethtool_stats_dict(self.phdl)