CI failure for CPU runners for regression tests

[jainapurva]

The CPU regression testing job has multiple failures with the issue:

ImportError: /opt/conda/lib/libstdc++.so.6: version `GLIBCXX_3.4.32' not found (required by /tmp/torchinductor_root/ez/cezzpgmr3zep66cb2qrfetykswkokhuzzyomho4jcfeuhfkzttox.so)

One sample failure

______ TestOptim.test_optim_smoke_optim_name_Adam8bit_bfloat16_device_cpu ______
  self = <test_low_bit_optim.TestOptim testMethod=test_optim_smoke_optim_name_Adam8bit_bfloat16_device_cpu>
  optim_name = 'Adam8bit', dtype = torch.bfloat16, device = 'cpu'
      @parametrize(
          "optim_name",
          ["Adam8bit", "AdamW8bit", "Adam4bit", "AdamW4bit", "AdamFp8", "AdamWFp8"],
      )
      @parametrize("dtype", [torch.float32, torch.bfloat16])
      @parametrize("device", _DEVICES)
      @skip_if_rocm("ROCm enablement in progress")
      def test_optim_smoke(self, optim_name, dtype, device):
          if optim_name.endswith("Fp8") and device == "cuda":
              if torch.cuda.get_device_capability() < (8, 9):
                  pytest.skip("FP8 CUDA requires compute capability >= 8.9")
      
          model = nn.Sequential(nn.Linear(32, 256), nn.ReLU(), nn.Linear(256, 32))
          model.to(device=device, dtype=dtype)
          optimizer = getattr(optim, optim_name)(model.parameters())
      
          x = torch.randn(4, 32, device=device, dtype=dtype)
          loss = model(x).sum()
          loss.backward()
  >       optimizer.step()
  test/test_low_bit_optim.py:143: 
  _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/optim/optimizer.py:493: in wrapper
      out = func(*args, **kwargs)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/utils/_contextlib.py:116: in decorate_context
      return func(*args, **kwargs)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torchao/optim/adam.py:142: in step
      torch.compile(single_param_adam, fullgraph=True, dynamic=False)(
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py:574: in _fn
      return fn(*args, **kwargs)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:1380: in __call__
      return self._torchdynamo_orig_callable(
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:547: in __call__
      return _compile(
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:986: in _compile
      guarded_code = compile_inner(code, one_graph, hooks, transform)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:715: in compile_inner
      return _compile_inner(code, one_graph, hooks, transform)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_utils_internal.py:95: in wrapper_function
      return function(*args, **kwargs)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:750: in _compile_inner
      out_code = transform_code_object(code, transform)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/bytecode_transformation.py:1361: in transform_code_object
      transformations(instructions, code_options)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:231: in _fn
      return fn(*args, **kwargs)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py:662: in transform
      tracer.run()
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py:2868: in run
      super().run()
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py:1052: in run
      while self.step():
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py:962: in step
      self.dispatch_table[inst.opcode](self, inst)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py:3048: in RETURN_VALUE
      self._return(inst)
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py:3033: in _return
      self.output.compile_subgraph(
  /opt/conda/envs/venv/lib/python3.10/site-packages/torch/_dynamo/output_graph.py:1136: in compile_subgraph
  Traceback (most recent call last):
    File "/home/ec2-user/actions-runner/_work/ao/ao/test-infra/.github/scripts/run_with_env_secrets.py", line 102, in <module>
  ImportError: /opt/conda/lib/libstdc++.so.6: version `GLIBCXX_3.4.32' not found (required by /tmp/torchinductor_root/ez/cezzpgmr3zep66cb2qrfetykswkokhuzzyomho4jcfeuhfkzttox.so)
  Set TORCH_LOGS="+dynamo" and TORCHDYNAMO_VERBOSE=1 for more information
  You can suppress this exception and fall back to eager by setting:
      import torch._dynamo
      torch._dynamo.config.suppress_errors = True
  To execute this test, run the following from the base repo dir:
      python test/test_low_bit_optim.py TestOptim.test_optim_smoke_optim_name_AdamWFp8_bfloat16_device_cpu
  This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0

[jainapurva]

The issue was caused due to update in test-infra docker image: pytorch/pytorch#167018
Issue fixed with: #3327 #3326