after update codes, i got error

[hsoftxl]

STEP 5: Training is progress

INFO: Debug mode ON. Logs will be saved to:
/Users/mac1/projects/source/nanowakeword/examples/trained_models/hope_A_v1/training_artifac
ts/training_debug/training_debug.log
INFO: Checkpointing is ENABLED. A checkpoint will be saved every 1000 steps.
INFO: Early stopping is DISABLED. Training will run for the full duration of 'steps'.
Traceback (most recent call last):
File "/opt/anaconda3/envs/wakeword/bin/nanowakeword-train", line 7, in
sys.exit(train())
File "/opt/anaconda3/envs/wakeword/lib/python3.10/site-packages/nanowakeword/trainer.py", line 1700, in train
best_model = nww.auto_train(
File "/opt/anaconda3/envs/wakeword/lib/python3.10/site-packages/nanowakeword/trainer.py", line 447, in auto_train
self.train_model(
File "/opt/anaconda3/envs/wakeword/lib/python3.10/site-packages/nanowakeword/trainer.py", line 663, in train_model
data_iterator = iter(X)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 494, in iter
return self._get_iterator()
File "/opt/anaconda3/envs/wakeword/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 427, in _get_iterator
return _MultiProcessingDataLoaderIter(self)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1172, in init
w.start()
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/process.py", line 121, in start
self._popen = self._Popen(self)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/context.py", line 224, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/context.py", line 288, in _Popen
return Popen(process_obj)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/popen_spawn_posix.py", line 32, in init
super().init(process_obj)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/popen_fork.py", line 19, in init
self._launch(process_obj)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/popen_spawn_posix.py", line 47, in _launch
reduction.dump(process_obj, fp)
File "/opt/anaconda3/envs/wakeword/lib/python3.10/multiprocessing/reduction.py", line 60, in dump
ForkingPickler(file, protocol).dump(obj)
AttributeError: Can't pickle local object 'train..IterDataset'

[muhammadabidPro]

⚠️ DataLoader Workers Note

By default, num_workers is set to 2.

On some systems, training may fail due to a PyTorch multiprocessing (pickling) issue.
If you encounter such an error, please set num_workers to 1 or 0 in your YAML config:

num_workers: 0

or

num_workers: 1

✅ We have tested this on Windows, Python v3.13.5 and training works correctly with the default setting.

Recommendation:

• If you see multiprocessing / pickling errors → use num_workers: 0
• Otherwise, the default value should work fine

[hsoftxl]

@muhammadabidPro worked

[alipas777-afk]

I got same error after update nanoWW

Starting NanoWakeWord training...

---

| \ | | \ \ / / | | \ \ / / | |
| | | __ _ _ __ \ \ /\ / / | | \ \ /\ / / _ __ _| |
| . |/ _ | ' \ / _ \ / / / | |/ / _ \ \/ \/ / _ \| '__/ _ |
| |\ | (| | | | | () \ /\ / (| | < __/\ /\ / () | | | (| |
|| _|_,|| ||_/ / / _,|_|__| / / __/|| _,_|

STEP 8: Verifying and Preprocessing Data Directories

INFO: 'generated_positive' already verified. Skipping.
INFO: 'generated_negative' already verified. Skipping.
INFO: 'background_noise' already verified. Skipping.
INFO: 'rir' already verified. Skipping.
INFO: Data verification and preprocessing complete.

INFO: Determining hardware-specific configurations...
STEP 9: Activating Intelligent Configuration Engine

Analyzing dataset characteristics...
Analyzing generated_positive files: 100%|██████████| 3208/3208 [00:08<00:00, 382.53it/s]
Analyzing generated_negative files: 100%|██████████| 53/53 [00:00<00:00, 383.34it/s]
Analyzing background_noise files: 100%|██████████| 1000/1000 [00:11<00:00, 88.40it/s]Analysis complete!

                    Dataset Statistics                         

┌───────────────┬─────────────────────────────────────────────────┐
│ Parameter │ Value │
├───────────────┼─────────────────────────────────────────────────┤
│ H_pos │ 0.5274278472222285 │
│ H_neg │ 0.010525868055555552 │
│ H_noise_paths │ {'./data/background_noise': 1.3888888888888888} │
│ H_noise │ 1.3888888888888888 │
│ A_noise │ 0.07119006916705985 │
│ N_rir │ 270 │
└───────────────┴─────────────────────────────────────────────────┘
INFO: Project assets will be saved in: /content/trained_models/balha_wakeword_final
INFO: Autotuning optimal clip duration...
INFO: Optimal clip duration autotuned to: 32000 samples (2.00 seconds).
INFO: Feature generation is disabled as 'transform_clips' is false and '--transform_clips' flag is not set.
INFO: Dataset initialized | Positives: 38496 | Negatives: 4636 | Total: 43132
INFO: Input Shape Detected: torch.Size([16, 96]) (1.28s context)
INFO: Initializing Neural Architecture...
INFO:
[WARNING] The 'CRNN' architecture is highly sensitive to hyperparameters and may exhibit convergence instability.

INFO: Using optimizer: ADAMW
INFO: Setting up learning rate scheduler: ONECYCLE
STEP 10: Training is progress

INFO: Debug mode ON. Logs will be saved to:
/content/trained_models/balha_wakeword_final/training_artifacts/training_debug/training_debug.log
INFO: Checkpointing is ENABLED. A checkpoint will be saved every 1000 steps.
INFO: Training for 20000 steps. Model checkpointing and early stopping will activate after 3000 warm-up steps.
Effective Training Configuration
┌─────────────────────────────────────────────────────────┬───────────────────────────────────────────────────────┐
│ Parameter │ Value │
├─────────────────────────────────────────────────────────┼───────────────────────────────────────────────────────┤
│ WARMUP_STEPS │ 3000 │
│ activation_function │ relu │
│ audio_processing.autotune_length.duration_buffer_ms │ 750 │
│ audio_processing.autotune_length.enabled │ True │
│ audio_processing.autotune_length.min_allowable_length │ 32000 │
│ audio_processing.autotune_length.num_samples_to_inspect │ 50 │
│ audio_processing.autotune_length.snap_to_min_tolerance │ 4000 │
│ background_paths_duplication_rate │ [1] │
│ batch_size │ 64 │
│ checkpoint_averaging_top_k │ 5 │
│ checkpointing.enabled │ True │
│ checkpointing.interval_steps │ 1000 │
│ checkpointing.limit │ 3 │
│ crnn_cnn_channels │ [16, 32, 32] │
│ crnn_rnn_type │ lstm │
│ debug_mode │ True │
│ dropout_prob │ 0.2 │
│ ema_alpha │ 0.01 │
│ embedding_dim │ 64 │
│ feature_manifest.backgrounds.b │ ./trained_models/balha_wakeword_final/features/noise… │
│ feature_manifest.negatives.n │ ./trained_models/balha_wakeword_final/features/hard_… │
│ feature_manifest.targets.t │ ./trained_models/balha_wakeword_final/features/posit… │
│ filter_interval_steps │ 10 │
│ layer_size │ 96 │
│ learning_rate_max │ 0.0008 │
│ loss_threshold_for_easy │ 0.01 │
│ lr_scheduler_type │ onecycle │
│ min_delta │ 0.0001 │
│ model_name │ balha_wakeword_final │
│ model_type │ crnn │
│ momentum │ 0.9 │
│ n_blocks │ 3 │
│ num_workers │ 3 │
│ optimizer_type │ adamw │
│ overwrite │ True │
│ show_training_summary │ True │
│ steps │ 20000 │
│ total_length │ 32000 │
│ train_model │ True │
│ transform_clips │ False │
│ weight_decay │ 0.01 │
└─────────────────────────────────────────────────────────┴───────────────────────────────────────────────────────┘
Training: 0%| | 9/20000 [00:01<1:01:10, 5.45it/s]

An error occurred during training: indices should be either on cpu or on the same device as the indexed tensor (cpu)

[muhammadabidPro]

@alipas777-afk,
I fixed it, Please try again and let me know the results.
By the way, this problem did not occur on my system.

[muhammadabidPro]

@alipas777-afk,
Another thing is why are you using the CRNN architecture? If it's just a test, then do it.
NOTE: This type of architecture did not give good results according to the previous code.
We tested DNN, RNN....https://github.com/arcosoph/nanowakeword/blob/main/STATUS.md