Upload folder using huggingface_hub

ISSUES.md

@@ -78,6 +78,23 @@ Let PyTorch Lightning do its job. Either:
    --trainer.default_root_dir uk_UA-ASMR/output
    ```
 
+## 4. GuardOnDataDependentSymNode: Could not guard on data-dependent expression
+
+The error you are encountering (`GuardOnDataDependentSymNode: Could not guard on data-dependent expression`) is caused by a recent change in PyTorch. In newer versions (like PyTorch 2.6+), `torch.onnx.export` defaults to the new strict TorchDynamo-based exporter. TorchDynamo crashes when it tries to trace data-dependent assertions, such as the assert `(discriminant >= 0).all()` runtime check located at line 174 of Piper's `transforms.py`.
+
+1. Open the file `/workspace/piper1-gpl/src/piper/train/export_onnx.py`
+2. Locate the `torch.onnx.export(` call around line 92.
+3. Add `dynamo=False`, to the list of arguments:
+
+```python
+torch.onnx.export(
+   ...
+   dynamo=False,          # <--- ADD THIS LINE
+   verbose=False,
+   ...
+)
+```
+
 <details>
 <summary>Usage</summary>
 

README.md

@@ -49,7 +49,7 @@ python3 setup.py build_ext --inplace
 
 # CRITICAL FIX for custom text phonemes in Piper `OHF-voice/piper1-gpl` (fixed in `kontextox/piper1-gpl`):
 # This patches dataset.py to properly use the custom phoneme map loaded via --data.phonemes_path
-sed -i 's/phonemes_to_ids(sentence_phonemes)/phonemes_to_ids(sentence_phonemes, id_map=self.piper_config.phoneme_id_map)/g' src/piper/train/vits/dataset.py
+# sed -i 's/phonemes_to_ids(sentence_phonemes)/phonemes_to_ids(sentence_phonemes, id_map=self.piper_config.phoneme_id_map)/g' src/piper/train/vits/dataset.py
 ```
 
 ```bash
@@ -116,9 +116,10 @@ python3 -m piper.train fit \
 
 _**Note**: `--trainer.default_root_dir` ensures PyTorch Lightning saves logs and checkpoints cleanly to `uk_UA-ASMR/output/lightning_logs/`_
 
-_**Note**: The NVIDIA driver on your system is too old (found version 12080): `pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118`_
+_**Note**: The NVIDIA driver on your system is too old (found version 12080) or NVIDIA GeForce RTX 5090 with CUDA capability `sm_120` is not compatible with the current PyTorch installation:_
 
-_**Note**: NVIDIA GeForce RTX 5090 with CUDA capability `sm_120` is not compatible with the current PyTorch installation: `pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128` for check `python -c "import torch; print(torch.__version__); print(torch.cuda.get_arch_list()); print(torch.randn(1).cuda())"`_
+- Check: `python -c "import torch; print(torch.__version__); print(torch.cuda.get_arch_list()); print(torch.randn(1).cuda())"`
+- Run: `pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128`
 
 _**Note**: Check CPU process `find uk_UA-ASMR/cache -name "*.pt" | wc -l`_