training_scripts/export_jit.py

#!/usr/bin/env python3
"""Export Piper checkpoint using JIT tracing"""

import sys
import torch
from pathlib import Path

sys.path.insert(0, str(Path("/root/piper_msa/piper1-gpl/src")))

from piper.train.vits.lightning import VitsModel

def main():
    checkpoint_path = "/root/piper_msa/piper1-gpl/lightning_logs/version_1/checkpoints/epoch=54-step=143440.ckpt"
    output_path = "/root/piper_msa/output/saudi_msa.onnx"
    
    print(f"Loading checkpoint: {checkpoint_path}")
    model = VitsModel.load_from_checkpoint(checkpoint_path, map_location="cpu")
    model_g = model.model_g
    
    # Inference only
    model_g.eval()
    
    with torch.no_grad():
        model_g.dec.remove_weight_norm()
    
    def infer_forward(text, text_lengths, scales, sid=None):
        noise_scale = scales[0]
        length_scale = scales[1]
        noise_scale_w = scales[2]
        audio = model_g.infer(
            text,
            text_lengths,
            noise_scale=noise_scale,
            length_scale=length_scale,
            noise_scale_w=noise_scale_w,
            sid=sid,
        )[0].unsqueeze(1)
        return audio
    
    model_g.forward = infer_forward
    
    num_symbols = model_g.n_vocab
    num_speakers = model_g.n_speakers
    
    dummy_input_length = 50
    sequences = torch.randint(
        low=0, high=num_symbols, size=(1, dummy_input_length), dtype=torch.long
    )
    sequence_lengths = torch.LongTensor([sequences.size(1)])
    
    sid = None
    if num_speakers > 1:
        sid = torch.LongTensor([0])
    
    scales = torch.FloatTensor([0.667, 1.0, 0.8])
    dummy_input = (sequences, sequence_lengths, scales, sid)
    
    print(f"Exporting to ONNX using JIT: {output_path}")
    
    # Use JIT tracing with legacy exporter
    with torch.no_grad():
        torch.onnx.export(
            model=model_g,
            args=dummy_input,
            f=output_path,
            verbose=False,
            opset_version=15,
            input_names=["input", "input_lengths", "scales", "sid"],
            output_names=["output"],
            dynamic_axes={
                "input": {0: "batch_size", 1: "phonemes"},
                "input_lengths": {0: "batch_size"},
                "output": {0: "batch_size", 2: "time"},
            },
            export_params=True,
            do_constant_folding=True,
            # Use legacy JIT-based exporter
            dynamo=False,
        )
    
    print(f"✓ Model exported successfully to: {output_path}")
    print(f"\nTo test the model:")
    print(f"  echo 'مرحبا بك' | piper --model {output_path} --config /root/piper_msa/output/config.json --output_file test.wav")

if __name__ == "__main__":
    main()