chat_nitro.py

import json
import torch
from torch.utils.data import DataLoader, Dataset
import torch.nn as nn

# Initialize tokenizer
class CustomTokenizer:
    def __init__(self, vocab):
        self.vocab = vocab
    
    def encode(self, text):
        tokens = text.split()
        ids = [self.vocab.get(token, self.vocab['[UNK]']) for token in tokens]
        return ids
    
    def decode(self, ids):
        tokens = [list(self.vocab.keys())[id] for id in ids if id != self.vocab['[PAD]'] and id < len(self.vocab)]
        return ' '.join(tokens)
    
    def pad_sequence(self, sequence, max_length):
        if len(sequence) < max_length:
            sequence = sequence + [self.vocab['[PAD]']] * (max_length - len(sequence))
        else:
            sequence = sequence[:max_length]
        return sequence

# Sample language model class
class LanguageModel(nn.Module):
    def __init__(self, vocab_size, embed_size, hidden_size):
        super(LanguageModel, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embed_size)
        self.rnn = nn.GRU(embed_size, hidden_size, batch_first=True)
        self.fc = nn.Linear(hidden_size, vocab_size)
    
    def forward(self, x, hidden=None):
        embedded = self.embedding(x)
        output, hidden = self.rnn(embedded, hidden)
        output = self.fc(output)
        return output, hidden

# Load the vocab from the JSON file
with open('vocab2.json', 'r') as f:
    vocab = json.load(f)

special_tokens = ['[PAD]', '[UNK]']
for token in special_tokens:
    if token not in vocab:
        vocab[token] = len(vocab)

tokenizer = CustomTokenizer(vocab)

# Model parameters
embed_size = 900
hidden_size = 900
vocab_size = max(vocab.values()) + 1

# Load the model
model = LanguageModel(vocab_size, embed_size, hidden_size)
model.load_state_dict(torch.load('language_model.nit'))
model.eval()

def generate_response(input_text, model, tokenizer, max_length=1000):
    encoded_input = tokenizer.encode(input_text)
    padded_input = tokenizer.pad_sequence(encoded_input, max_length)
    input_tensor = torch.tensor(padded_input).unsqueeze(0)  # Add batch dimension
    
    with torch.no_grad():
        outputs, _ = model(input_tensor)
    
    predicted_ids = torch.argmax(outputs, dim=2).squeeze().tolist()
    predicted_text = tokenizer.decode(predicted_ids)
    
    return predicted_text

# Test the model with a new text
while True:
    test_text = input(">>>")
    response = generate_response(test_text, model, tokenizer)
    print("Input:", test_text)
    print("Response:", response)