Delete train.py

train.py

@@ -1,69 +0,0 @@
-import torch
-import torch.nn as nn
-import torchaudio
-from torch.utils.data import DataLoader
-from datasets import load_dataset
-from model.model import RealtimeTTS
-from model.config import TTSConfig
-from model.tokenizer import TTSTokenizer
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-config = TTSConfig()
-
-# Load tokenizer
-tokenizer = TTSTokenizer("tts_tokenizer.model")
-
-# Load dataset
-dataset = load_dataset("csv", data_files={"train": "train.csv"})["train"]
-
-mel_transform = torchaudio.transforms.MelSpectrogram(
-    sample_rate=22050,
-    n_mels=config.mel_bins
-)
-
-def preprocess(example):
-    audio, sr = torchaudio.load(example["audio_path"])
-    mel = mel_transform(audio).transpose(1, 2)
-    tokens = tokenizer.encode(example["text"])
-
-    return {
-        "tokens": torch.tensor(tokens),
-        "mel": mel.squeeze(0)
-    }
-
-dataset = dataset.map(preprocess)
-
-def collate_fn(batch):
-    tokens = [item["tokens"] for item in batch]
-    mels = [item["mel"] for item in batch]
-
-    tokens = nn.utils.rnn.pad_sequence(tokens, batch_first=True)
-    mels = nn.utils.rnn.pad_sequence(mels, batch_first=True)
-
-    return tokens, mels
-
-dataloader = DataLoader(dataset, batch_size=8, shuffle=True, collate_fn=collate_fn)
-
-model = RealtimeTTS(config).to(device)
-optimizer = torch.optim.AdamW(model.parameters(), lr=3e-4)
-loss_fn = nn.MSELoss()
-
-# Training loop
-for epoch in range(10):
-    model.train()
-    for tokens, mels in dataloader:
-        tokens = tokens.to(device)
-        mels = mels.to(device)
-
-        mel_input = torch.zeros_like(mels)
-        output = model(tokens, mel_input)
-
-        loss = loss_fn(output, mels)
-
-        optimizer.zero_grad()
-        loss.backward()
-        optimizer.step()
-
-    print(f"Epoch {epoch} Loss: {loss.item()}")
-
-torch.save(model.state_dict(), "model.pt")