I encountered a version issue while running PyTorch. The code runs smoothly with PyTorch 1.6, but it doesn't work with PyTorch 1.13.

vision
1

The code that is causing the error is as follows，Does anyone know how to resolve this issue without downgrading PyTorch? My GPU is a 4060, and I’ve tried lower versions of PyTorch, but they don’t work correctly.

Traceback (most recent call last):
File “D:\桌面\SA-MASAC-master\train_masac.py”, line 158, in
main(args)
File “D:\桌面\SA-MASAC-master\train_masac.py”, line 152, in main
ddpg_continuous(config)
File “D:\桌面\SA-MASAC-master\train_masac.py”, line 115, in ddpg_continuous
config.mini_batch_size, config.alpha)
File “D:\桌面\SA-MASAC-master\robust_masac.py”, line 576, in init
self.actor = RobustActorNet(obs_dim, hid_shape, act_dim).to(config.DEVICE)
File “D:\桌面\SA-MASAC-master\robust_masac.py”, line 254, in init
self.a_net = BoundedModule(self.a_net, (torch.empty(size=(1, state_dim)),), device=Config.DEVICE)
File “D:\桌面\SA-MASAC-master\auto_LiRPA\bound_general.py”, line 35, in init
self._convert(model, global_input)
File “D:\桌面\SA-MASAC-master\auto_LiRPA\bound_general.py”, line 459, in _convert
nodesOP, nodesIn, nodesOut, template = self._convert_nodes(model, global_input)
File “D:\桌面\SA-MASAC-master\auto_LiRPA\bound_general.py”, line 296, in _convert_nodes
nodesOP, nodesIn, nodesOut, template = parse_module(model, global_input_cpu)
File “D:\桌面\SA-MASAC-master\auto_LiRPA\parse_graph.py”, line 170, in parse_module
trace_graph = torch.onnx.optimize_trace(trace, torch.onnx.OperatorExportTypes.ONNX)
File "D:\anaconda3\envs\PT\lib\site-packages\torch\onnx_init.py", line 394, in _optimize_trace
return utils._optimize_graph(graph, operator_export_type)
File “D:\anaconda3\envs\PT\lib\site-packages\torch\onnx\utils.py”, line 278, in _optimize_graph
graph, params_dict, symbolic_helper.is_caffe2_aten_fallback()
TypeError: _jit_pass_onnx_unpack_quantized_weights(): incompatible function arguments. The following argument types are supported:
1. (arg0: torch::jit::Graph, arg1: Dict[str, IValue], arg2: bool) → Dict[str, IValue]

2

TorchScript is in maintenance mode, so remove scripting or tracing from your model and try to use torch.compile instead in the recent PyTorch release.

3

hello

I also encountered the same problem, have you solved it?

TypeError: _jit_pass_onnx_unpack_quantized_weights(): incompatible function arguments. The following argument types are supported:
1. (arg0: torch::jit::Graph, arg1: Dict[str, IValue], arg2: bool) → Dict[str, IValue]

4

@cxy22 have you solved it? also same problems here.