pytorch/slm_train.py

import torch
import torch.nn as nn
import torch.optim as optim

# 1. Prepare the Dataset

def prepare_dataset(filepath, seq_length):
    """
    Prepares the dataset for training from a text file.

    Args:
        filepath (str): Path to the text file (e.g., 'dataset.txt').
        seq_length (int): The length of input sequences.

    Returns:
        tuple: vocab (set), char_to_index (dict), index_to_char (dict),
               input_sequences (list), target_sequences (list)
    """
    try:
        with open(filepath, 'r', encoding='utf-8') as file:
            text = file.read()
    except FileNotFoundError:
        print(f"Error: File '{filepath}' not found. Make sure the file exists in the correct directory.")
        return None, None, None, None, None

    vocab = sorted(list(set(text)))
    char_to_index = {char: index for index, char in enumerate(vocab)}
    index_to_char = {index: char for index, char in enumerate(vocab)}

    input_sequences = []
    target_sequences = []

    for i in range(0, len(text) - seq_length):
        input_seq = text[i:i + seq_length]
        target_seq = text[i + seq_length]
        input_sequences.append([char_to_index[char] for char in input_seq])
        target_sequences.append(char_to_index[target_seq])

    return vocab, char_to_index, index_to_char, input_sequences, target_sequences


# 2. Define the Language Model (Simple RNN)

class SimpleRNNLM(nn.Module):
    def __init__(self, vocab_size, embedding_dim, hidden_dim, num_layers):
        super(SimpleRNNLM, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embedding_dim)
        self.rnn = nn.RNN(embedding_dim, hidden_dim, num_layers, batch_first=True)
        self.fc = nn.Linear(hidden_dim, vocab_size)

    def forward(self, input_seq, hidden):
        embedded = self.embedding(input_seq)
        output, hidden = self.rnn(embedded, hidden)
        output = self.fc(output[:, -1, :])
        return output, hidden

    def init_hidden(self, batch_size, num_layers, hidden_dim):
        return torch.zeros(num_layers, batch_size, hidden_dim)


# Example Usage
dataset_filepath = 'dataset.txt'  # Path to your dataset text file
seq_length = 64

vocab, char_to_index, index_to_char, input_seqs, target_seqs = prepare_dataset(dataset_filepath, seq_length)

if vocab is None:
    exit()

print(f"Vocabulary Size: {len(vocab)}")
print(f"Number of Input Sequences: {len(input_seqs)}")


# 3. Instantiate Model, Loss Function, and Optimizer

vocab_size = len(vocab)
embedding_dim = 32
hidden_dim = 64
num_layers = 1
learning_rate = 0.01
num_epochs = 10

model = SimpleRNNLM(vocab_size, embedding_dim, hidden_dim, num_layers)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

device = torch.device("cpu")
model.to(device)
criterion.to(device)


# 4. Training Loop

batch_size = 256

for epoch in range(num_epochs):
    model.train()
    total_loss = 0

    for i in range(0, len(input_seqs), batch_size):
        input_batch = input_seqs[i:i+batch_size]
        target_batch = target_seqs[i:i+batch_size]

        input_batch_tensor = torch.LongTensor(input_batch).to(device)
        target_batch_tensor = torch.LongTensor(target_batch).to(device)

        hidden = model.init_hidden(len(input_batch), num_layers, hidden_dim).to(device)

        optimizer.zero_grad()

        output, hidden = model(input_batch_tensor, hidden)
        loss = criterion(output, target_batch_tensor)

        loss.backward()
        optimizer.step()

        total_loss += loss.item()

    average_loss = total_loss / (len(input_seqs) // batch_size + (len(input_seqs) % batch_size != 0))
    print(f"Epoch [{epoch+1}/{num_epochs}], Loss: {average_loss:.4f}")


# 5. Text Generation

def generate_text(model, start_text, predict_len, char_to_index, index_to_char, vocab, device):
    model.eval()
    generated_text = start_text

    input_sequence = [char_to_index[char] for char in start_text]
    input_tensor = torch.LongTensor([input_sequence]).to(device)

    hidden = model.init_hidden(1, num_layers, hidden_dim).to(device)

    with torch.no_grad():
        for _ in range(predict_len):
            output, hidden = model(input_tensor, hidden)

            probabilities = torch.softmax(output, dim=1)
            predicted_index = torch.multinomial(probabilities, 1).item()
            predicted_char = index_to_char[predicted_index]

            generated_text += predicted_char

            input_sequence = input_sequence[1:] + [predicted_index]
            input_tensor = torch.LongTensor([input_sequence]).to(device)

    return generated_text


# Example Generation
start_text = "The "
predict_length = 500

generated_output = generate_text(model, start_text, predict_length, char_to_index, index_to_char, vocab, device)
print("\nGenerated Text:")
print(generated_output)