Error when launching training within the notebook nuplan_framework.ipynb

[CBeaune]

Hello, I'm a newcomer in the nuplan framework so I apologize if these are simple issues, but I did not find any closed issues involving these.
I'm trying to run the tutorials starting by the nuplan_framework.ipynb but i get the following issue when trying to launch the training within the notebook, it seems related to the loading of the pretrained model:

AttributeError: Error instantiating 'nuplan.planning.training.modeling.models.raster_model.RasterModel' : module 'torch' has no attribute 'frombuffer'

It seems that the torch attribute is not reachable. Is it an issue with the torch version? Mine is '1.9.0+cu111' as written in the requirements_torch.txt of the devkit.

Any help is welcome! Thanks

[ZhaoYangbjtu]

I have same problem whenI use version nuplan-devkit-v1.2-release . All the installation steps are successful. stack trace is below; seem like another lib safetensors need newer version of torch. Is there anything wrong inside requirments.txt or requirments_torch.txt?

---

AttributeError Traceback (most recent call last) File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py:62, in _call_target(target, *args, *kwargs) 60 v._set_parent(None) ---> 62 return target(args, **kwargs) 63 except Exception as e:

File ~/workspace/nuplan-devkit/nuplan/planning/training/modeling/models/raster_model.py:58, in RasterModel.init(self, feature_builders, target_builders, model_name, pretrained, num_input_channels, num_features_per_pose, future_trajectory_sampling) 57 num_output_features = future_trajectory_sampling.num_poses * num_features_per_pose ---> 58 self._model = timm.create_model(model_name, pretrained=pretrained, num_classes=0, in_chans=num_input_channels) 59 mlp = torch.nn.Linear(in_features=self._model.num_features, out_features=num_output_features)

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_factory.py:114, in create_model(model_name, pretrained, pretrained_cfg, pretrained_cfg_overlay, checkpoint_path, scriptable, exportable, no_jit, kwargs) 113 with set_layer_config(scriptable=scriptable, exportable=exportable, no_jit=no_jit): --> 114 model = create_fn( 115 pretrained=pretrained, 116 pretrained_cfg=pretrained_cfg, 117 pretrained_cfg_overlay=pretrained_cfg_overlay, 118 kwargs, 119 ) 121 if checkpoint_path:

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/resnet.py:1276, in resnet50(pretrained, kwargs) 1275 model_args = dict(block=Bottleneck, layers=[3, 4, 6, 3], kwargs) -> 1276 return _create_resnet('resnet50', pretrained, dict(model_args, kwargs))

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/resnet.py:547, in _create_resnet(variant, pretrained, kwargs) 546 def _create_resnet(variant, pretrained=False, kwargs): --> 547 return build_model_with_cfg(ResNet, variant, pretrained, **kwargs)

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_builder.py:393, in build_model_with_cfg(model_cls, variant, pretrained, pretrained_cfg, pretrained_cfg_overlay, model_cfg, feature_cfg, pretrained_strict, pretrained_filter_fn, kwargs_filter, **kwargs) 392 if pretrained: --> 393 load_pretrained( 394 model, 395 pretrained_cfg=pretrained_cfg, 396 num_classes=num_classes_pretrained, 397 in_chans=kwargs.get('in_chans', 3), 398 filter_fn=pretrained_filter_fn, 399 strict=pretrained_strict, 400 ) 402 # Wrap the model in a feature extraction module if enabled

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_builder.py:186, in load_pretrained(model, pretrained_cfg, num_classes, in_chans, filter_fn, strict) 185 else: --> 186 state_dict = load_state_dict_from_hf(pretrained_loc) 187 else:

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_hub.py:183, in load_state_dict_from_hf(model_id, filename) 180 _logger.info( 181 f"[{model_id}] Safe alternative available for '{filename}' " 182 f"(as '{safe_filename}'). Loading weights using safetensors.") --> 183 return safetensors.torch.load_file(cached_safe_file, device="cpu") 184 except EntryNotFoundError:

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/safetensors/torch.py:261, in load_file(filename, device) 260 for k in f.keys(): --> 261 result[k] = f.get_tensor(k) 262 return result

AttributeError: module 'torch' has no attribute 'frombuffer'

During handling of the above exception, another exception occurred:

AttributeError Traceback (most recent call last) Cell In[6], line 4 1 from nuplan.planning.script.run_training import main as main_train 3 # Run the training loop, optionally inspect training artifacts through tensorboard (above cell) ----> 4 main_train(cfg)

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/hydra/main.py:44, in main..main_decorator..decorated_main(cfg_passthrough) 41 @functools.wraps(task_function) 42 def decorated_main(cfg_passthrough: Optional[DictConfig] = None) -> Any: 43 if cfg_passthrough is not None: ---> 44 return task_function(cfg_passthrough) 45 else: 46 args = get_args_parser()

File ~/workspace/nuplan-devkit/nuplan/planning/script/run_training.py:59, in main(cfg) 56 if cfg.py_func == 'train': 57 # Build training engine 58 with ProfilerContextManager(cfg.output_dir, cfg.enable_profiling, "build_training_engine"): ---> 59 engine = build_training_engine(cfg, worker) 61 # Run training 62 logger.info('Starting training...')

File ~/workspace/nuplan-devkit/nuplan/planning/training/experiments/training.py:44, in build_training_engine(cfg, worker) 41 logger.info('Building training engine...') 43 # Create model ---> 44 torch_module_wrapper = build_torch_module_wrapper(cfg.model) 46 # Build the datamodule 47 datamodule = build_lightning_datamodule(cfg, worker, torch_module_wrapper)

File ~/workspace/nuplan-devkit/nuplan/planning/script/builders/model_builder.py:19, in build_torch_module_wrapper(cfg) 13 """ 14 Builds the NN module. 15 :param cfg: DictConfig. Configuration that is used to run the experiment. 16 :return: Instance of TorchModuleWrapper. 17 """ 18 logger.info('Building TorchModuleWrapper...') ---> 19 model = instantiate(cfg) 20 validate_type(model, TorchModuleWrapper) 21 logger.info('Building TorchModuleWrapper...DONE!')

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py:180, in instantiate(config, *args, *kwargs) 177 recursive = config.pop(_Keys.RECURSIVE, True) 178 convert = config.pop(_Keys.CONVERT, ConvertMode.NONE) --> 180 return instantiate_node(config, args, recursive=recursive, convert=convert) 181 else: 182 raise InstantiationException( 183 "Top level config has to be OmegaConf DictConfig, plain dict, or a Structured Config class or instance" 184 )

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py:249, in instantiate_node(node, convert, recursive, args) 245 value = instantiate_node( 246 value, convert=convert, recursive=recursive 247 ) 248 kwargs[key] = _convert_node(value, convert) --> 249 return _call_target(target, args, **kwargs) 250 else: 251 # If ALL or PARTIAL non structured, instantiate in dict and resolve interpolations eagerly. 252 if convert == ConvertMode.ALL or ( 253 convert == ConvertMode.PARTIAL and node._metadata.object_type is None 254 ):

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py:64, in _call_target(target, *args, *kwargs) 62 return target(args, **kwargs) 63 except Exception as e: ---> 64 raise type(e)( 65 f"Error instantiating '{_convert_target_to_string(target)}' : {e}" 66 ).with_traceback(sys.exc_info()[2])

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py:62, in _call_target(target, *args, *kwargs) 59 if OmegaConf.is_config(v): 60 v._set_parent(None) ---> 62 return target(args, **kwargs) 63 except Exception as e: 64 raise type(e)( 65 f"Error instantiating '{_convert_target_to_string(target)}' : {e}" 66 ).with_traceback(sys.exc_info()[2])

File ~/workspace/nuplan-devkit/nuplan/planning/training/modeling/models/raster_model.py:58, in RasterModel.init(self, feature_builders, target_builders, model_name, pretrained, num_input_channels, num_features_per_pose, future_trajectory_sampling) 51 super().init( 52 feature_builders=feature_builders, 53 target_builders=target_builders, 54 future_trajectory_sampling=future_trajectory_sampling, 55 ) 57 num_output_features = future_trajectory_sampling.num_poses * num_features_per_pose ---> 58 self._model = timm.create_model(model_name, pretrained=pretrained, num_classes=0, in_chans=num_input_channels) 59 mlp = torch.nn.Linear(in_features=self._model.num_features, out_features=num_output_features) 61 if hasattr(self._model, 'classifier'):

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_factory.py:114, in create_model(model_name, pretrained, pretrained_cfg, pretrained_cfg_overlay, checkpoint_path, scriptable, exportable, no_jit, kwargs) 112 create_fn = model_entrypoint(model_name) 113 with set_layer_config(scriptable=scriptable, exportable=exportable, no_jit=no_jit): --> 114 model = create_fn( 115 pretrained=pretrained, 116 pretrained_cfg=pretrained_cfg, 117 pretrained_cfg_overlay=pretrained_cfg_overlay, 118 kwargs, 119 ) 121 if checkpoint_path: 122 load_checkpoint(model, checkpoint_path)

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/resnet.py:1276, in resnet50(pretrained, kwargs) 1273 """Constructs a ResNet-50 model. 1274 """ 1275 model_args = dict(block=Bottleneck, layers=[3, 4, 6, 3], kwargs) -> 1276 return _create_resnet('resnet50', pretrained, dict(model_args, kwargs))

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/resnet.py:547, in _create_resnet(variant, pretrained, kwargs) 546 def _create_resnet(variant, pretrained=False, kwargs): --> 547 return build_model_with_cfg(ResNet, variant, pretrained, **kwargs)

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_builder.py:393, in build_model_with_cfg(model_cls, variant, pretrained, pretrained_cfg, pretrained_cfg_overlay, model_cfg, feature_cfg, pretrained_strict, pretrained_filter_fn, kwargs_filter, **kwargs) 391 num_classes_pretrained = 0 if features else getattr(model, 'num_classes', kwargs.get('num_classes', 1000)) 392 if pretrained: --> 393 load_pretrained( 394 model, 395 pretrained_cfg=pretrained_cfg, 396 num_classes=num_classes_pretrained, 397 in_chans=kwargs.get('in_chans', 3), 398 filter_fn=pretrained_filter_fn, 399 strict=pretrained_strict, 400 ) 402 # Wrap the model in a feature extraction module if enabled 403 if features:

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_builder.py:186, in load_pretrained(model, pretrained_cfg, num_classes, in_chans, filter_fn, strict) 184 state_dict = load_state_dict_from_hf(*pretrained_loc) 185 else: --> 186 state_dict = load_state_dict_from_hf(pretrained_loc) 187 else: 188 model_name = pretrained_cfg.get('architecture', 'this model')

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/timm/models/_hub.py:183, in load_state_dict_from_hf(model_id, filename) 179 cached_safe_file = hf_hub_download(repo_id=hf_model_id, filename=safe_filename, revision=hf_revision) 180 _logger.info( 181 f"[{model_id}] Safe alternative available for '{filename}' " 182 f"(as '{safe_filename}'). Loading weights using safetensors.") --> 183 return safetensors.torch.load_file(cached_safe_file, device="cpu") 184 except EntryNotFoundError: 185 pass

File ~/miniconda3/envs/nuplan/lib/python3.9/site-packages/safetensors/torch.py:261, in load_file(filename, device) 259 with safe_open(filename, framework="pt", device=device) as f: 260 for k in f.keys(): --> 261 result[k] = f.get_tensor(k) 262 return result

AttributeError: Error instantiating 'nuplan.planning.training.modeling.models.raster_model.RasterModel' : module 'torch' has no attribute 'frombuffer'

[michael-motional]

Could you try locking timm==0.6.7 here and rebuild the environment from scratch -- we use pytorch 1.9 which I think isn't compatible with recent (or maybe any) versions of safetensor.

[CBeaune]

It worked when changing the required timm version! I had an issue with the train loss being nan after 1st epoch but resolved it as mentionned in #91 Thanks for the help !

[michael-motional]

Nice, we'll address whenever the next release is made