revert nemo-rl patch

dockerfiles/nemo-rl.patch

@@ -1,20 +1,33 @@
 diff --git a/nemo_rl/algorithms/sft.py b/nemo_rl/algorithms/sft.py
-index a447510..e987002 100644
+index 6de9ac81..fc232ac6 100644
 --- a/nemo_rl/algorithms/sft.py
 +++ b/nemo_rl/algorithms/sft.py
-@@ -138,6 +138,7 @@ def setup(
+@@ -137,6 +137,7 @@ def setup(
          shuffle=data_config["shuffle"],
          collate_fn=rl_collate_fn,
          drop_last=True,
 +        num_workers=10,
      )
 
      if last_checkpoint_path is not None:
-@@ -152,6 +153,7 @@ def setup(
+@@ -151,6 +152,7 @@ def setup(
          shuffle=False,
          collate_fn=rl_collate_fn,
-         drop_last=False,
+         drop_last=True,
 +        num_workers=10,
      )
 
      # ==========================
+diff --git a/nemo_rl/models/policy/megatron_policy_worker.py b/nemo_rl/models/policy/megatron_policy_worker.py
+index 10c8cf33..b5ab26f0 100644
+--- a/nemo_rl/models/policy/megatron_policy_worker.py
++++ b/nemo_rl/models/policy/megatron_policy_worker.py
+@@ -586,6 +586,8 @@ class MegatronPolicyWorker:
+                 "https://github.com/NVIDIA/Megatron-LM/blob/1ab876ddc4c1893c76f26d775226a8d1dcdfb3d2/megatron/core/transformer/mlp.py#L174."
+             )
+         model_cfg.apply_rope_fusion = self.cfg["megatron_cfg"]["apply_rope_fusion"]
++        if "layernorm_epsilon" in self.cfg["megatron_cfg"]:
++            model_cfg.layernorm_epsilon = self.cfg["megatron_cfg"]["layernorm_epsilon"]
+
+         checkpoint_config = CheckpointConfig(
+             save_interval=100,