fix: Set use_flashinfer_fused_rope to False (#1636)

Signed-off-by: Shanmugam Ramasamy <[email protected]>
Co-authored-by: Shanmugam Ramasamy <[email protected]>

Author: shanmugamr1992

nemo_rl/models/policy/workers/megatron_policy_worker.py

@@ -1850,7 +1850,7 @@ def generate(
                 "tensor_model_parallel_size"
             ],
             use_cuda_graphs_for_non_decode_steps=use_cuda_graphs_for_non_decode_steps,
-            use_flashinfer_fused_rope=None,
+            use_flashinfer_fused_rope=False,
             unified_memory_level=unified_memory_level,
             max_tokens_override=max_tokens,
         )

tests/unit/models/policy/test_megatron_worker.py

@@ -547,9 +547,6 @@ def generation_setup(request, tiny_llama_model_path):
             cluster.shutdown()
 
 
-@pytest.mark.skip(
-    reason="Skipping mcore generation test to unblock mcore bump. Current issue summarized in https://github.com/NVIDIA-NeMo/RL/issues/1633"
-)
 @pytest.mark.timeout(240)
 @pytest.mark.parametrize(
     "generation_setup",