model.py

# coding=utf-8
# Copyright 2025 The ACC Team Authors and The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""ACC-FiPhi-NeuralMark-V3"""
























import random
import math
import time
import os




PHI = (1 + math.sqrt(5)) / 2




text = os.getenv("FiPhi-NeuralMark-V3/TRAINING_DATA.txt")




words = text.split()




trigram_chain = {}
for i in range(len(words) - 2):
    key = (words[i], words[i + 1])
    next_word = words[i + 2]
    if key not in trigram_chain:
        trigram_chain[key] = []
    trigram_chain[key].append(next_word)








def generate_text(length):
    if len(words) < 2:
        return ""
    key = random.choice(list(trigram_chain.keys()))
    result = [key[0], key[1]]
    for _ in range(length - 2):
        if key in trigram_chain:
            next_word = random.choice(trigram_chain[key])
            result.append(next_word)
            key = (key[1], next_word)
        else:
            break
    return " ".join(result)








class NeuralNetwork:
    def __init__(self, input_size, hidden_size1, hidden_size2, output_size):
        self.input_size = input_size
        self.hidden_size1 = hidden_size1
        self.hidden_size2 = hidden_size2
        self.output_size = output_size
        self.weights_input_hidden1 = [
            [random.random() for _ in range(input_size)] for _ in range(hidden_size1)
        ]
        self.weights_hidden1_hidden2 = [
            [random.random() for _ in range(hidden_size1)] for _ in range(hidden_size2)
        ]
        self.weights_hidden2_output = [
            [random.random() for _ in range(hidden_size2)] for _ in range(output_size)
        ]
        self.bias_hidden1 = [random.random() for _ in range(hidden_size1)]
        self.bias_hidden2 = [random.random() for _ in range(hidden_size2)]
        self.bias_output = [random.random() for _ in range(output_size)]




    def sigmoid(self, x):
        return 1 / (1 + math.exp(-x))




    def sigmoid_derivative(self, x):
        return x * (1 - x)




    def forward(self, inputs):
        self.hidden_input1 = [
            sum(inputs[i] * self.weights_input_hidden1[j][i] for i in range(self.input_size)) + self.bias_hidden1[j]
            for j in range(self.hidden_size1)
        ]
        self.hidden_output1 = [self.sigmoid(x) for x in self.hidden_input1]
        self.hidden_input2 = [
            sum(self.hidden_output1[i] * self.weights_hidden1_hidden2[j][i] for i in range(self.hidden_size1)) + self.bias_hidden2[j]
            for j in range(self.hidden_size2)
        ]
        self.hidden_output2 = [self.sigmoid(x) for x in self.hidden_input2]
        self.output_input = [
            sum(self.hidden_output2[i] * self.weights_hidden2_output[j][i] for i in range(self.hidden_size2)) + self.bias_output[j]
            for j in range(self.output_size)
        ]
        self.output_output = [self.sigmoid(x) for x in self.output_input]
        return self.output_output




    def backward(self, inputs, target, learning_rate=0.1):
        output_errors = [target[i] - self.output_output[i] for i in range(self.output_size)]
        output_deltas = [output_errors[i] * self.sigmoid_derivative(self.output_output[i])
                           for i in range(self.output_size)]
        hidden2_errors = [
            sum(output_deltas[k] * self.weights_hidden2_output[k][j] for k in range(self.output_size))
            for j in range(self.hidden_size2)
        ]
        hidden2_deltas = [hidden2_errors[j] * self.sigmoid_derivative(self.hidden_output2[j])
                          for j in range(self.hidden_size2)]
        hidden1_errors = [
            sum(hidden2_deltas[k] * self.weights_hidden1_hidden2[k][j] for k in range(self.hidden_size2))
            for j in range(self.hidden_size1)
        ]
        hidden1_deltas = [hidden1_errors[j] * self.sigmoid_derivative(self.hidden_output1[j])
                          for j in range(self.hidden_size1)]




        for i in range(self.output_size):
            for j in range(self.hidden_size2):
                self.weights_hidden2_output[i][j] += learning_rate * output_deltas[i] * self.hidden_output2[j]
            self.bias_output[i] += learning_rate * output_deltas[i]




        for i in range(self.hidden_size2):
            for j in range(self.hidden_size1):
                self.weights_hidden1_hidden2[i][j] += learning_rate * hidden2_deltas[i] * self.hidden_output1[j]
            self.bias_hidden2[i] += learning_rate * hidden2_deltas[i]




        for i in range(self.hidden_size1):
            for j in range(self.input_size):
                self.weights_input_hidden1[i][j] += learning_rate * hidden1_deltas[i] * inputs[j]
            self.bias_hidden1[i] += learning_rate * hidden1_deltas[i]








class RecurrentNeuralNetwork:
    def __init__(self, input_size, hidden_size, output_size):
        self.input_size = input_size
        self.hidden_size = hidden_size
        self.output_size = output_size
        self.weights_input_hidden = [
            [random.random() for _ in range(input_size)] for _ in range(hidden_size)
        ]
        self.weights_hidden_hidden = [
            [random.random() for _ in range(hidden_size)] for _ in range(hidden_size)
        ]
        self.weights_hidden_output = [
            [random.random() for _ in range(hidden_size)] for _ in range(output_size)
        ]
        self.bias_hidden = [random.random() for _ in range(hidden_size)]
        self.bias_output = [random.random() for _ in range(output_size)]




    def sigmoid(self, x):
        return 1 / (1 + math.exp(-x))




    def sigmoid_derivative(self, x):
        return x * (1 - x)




    def forward(self, inputs):
        self.hidden_state = [0] * self.hidden_size
        for _ in range(2):
            for i in range(len(inputs)):
                current_input = [0] * self.input_size
                current_input[i] = inputs[i]
                combined = [
                    sum(current_input[k] * self.weights_input_hidden[j][k] for k in range(self.input_size)) +
                    sum(self.hidden_state[k] * self.weights_hidden_hidden[j][k] for k in range(self.hidden_size)) +
                    self.bias_hidden[j]
                    for j in range(self.hidden_size)
                ]
                self.hidden_state = [self.sigmoid(val) for val in combined]
        output = [
            sum(self.hidden_state[k] * self.weights_hidden_output[i][k] for k in range(self.hidden_size)) +
            self.bias_output[i]
            for i in range(self.output_size)
        ]
        return [self.sigmoid(o) for o in output]




    def backward(self, inputs, target, learning_rate=0.1):
        output = self.forward(inputs)
        output_errors = [target[i] - output[i] for i in range(self.output_size)]
        output_deltas = [output_errors[i] * self.sigmoid_derivative(output[i])
                           for i in range(self.output_size)]
        hidden_errors = [
            sum(output_deltas[k] * self.weights_hidden_output[k][j] for k in range(self.output_size))
            for j in range(self.hidden_size)
        ]
        hidden_deltas = [hidden_errors[j] * self.sigmoid_derivative(self.hidden_state[j])
                         for j in range(self.hidden_size)]




        for i in range(self.output_size):
            for j in range(self.hidden_size):
                self.weights_hidden_output[i][j] += learning_rate * output_deltas[i] * self.hidden_state[j]
            self.bias_output[i] += learning_rate * output_deltas[i]




        for j in range(self.hidden_size):
            for k in range(self.input_size):
                self.weights_input_hidden[j][k] += learning_rate * hidden_deltas[j] * (inputs[k] if k < len(inputs) else 0)
            self.bias_hidden[j] += learning_rate * hidden_deltas[j]
        return output_errors








class ConvolutionalNeuralNetwork:
    def __init__(self, input_length, kernel_size1, kernel_size2, output_size):
        self.input_length = input_length
        self.kernel_size1 = kernel_size1
        self.kernel_size2 = kernel_size2
        self.output_size = output_size
        self.kernel1 = [random.random() for _ in range(kernel_size1)]
        self.bias1 = random.random()
        self.kernel2 = [random.random() for _ in range(kernel_size2)]
        self.bias2 = random.random()
        self.weights_output = [
            [random.random() for _ in range(input_length - kernel_size1 - kernel_size2 + 2)]
            for _ in range(output_size)
        ]
        self.bias_output = [random.random() for _ in range(output_size)]




    def relu(self, x):
        return x if x > 0 else 0




    def relu_derivative(self, x):
        return 1 if x > 0 else 0




    def convolve(self, inputs, kernel, bias):
        conv_output = []
        kernel_size = len(kernel)
        for i in range(len(inputs) - kernel_size + 1):
            s = sum(inputs[i + j] * kernel[j] for j in range(kernel_size)) + bias
            conv_output.append(self.relu(s))
        return conv_output




    def forward(self, inputs):
        conv1 = self.convolve(inputs, self.kernel1, self.bias1)
        conv2 = self.convolve(conv1, self.kernel2, self.bias2)
        fc_input = conv2
        output = [
            sum(fc_input[j] * self.weights_output[i][j] for j in range(len(fc_input))) + self.bias_output[i]
            for i in range(self.output_size)
        ]
        return [self.relu(o) for o in output]




    def backward(self, inputs, target, learning_rate=0.1):
        output = self.forward(inputs)
        output_errors = [target[i] - output[i] for i in range(self.output_size)]
        for i in range(self.output_size):
            for j in range(len(inputs) - self.kernel_size1 - self.kernel_size2 + 2):
                self.weights_output[i][j] += learning_rate * output_errors[i]
            self.bias_output[i] += learning_rate * output_errors[i]
        return output_errors








class GeneticAlgorithm:
    def __init__(self, population_size, gene_length):
        self.population_size = population_size
        self.gene_length = gene_length
        self.population = [
            [random.random() for _ in range(gene_length)] for _ in range(population_size)
        ]




    def fitness(self, individual):
        return -sum((gene - PHI) ** 2 for gene in individual)




    def selection(self):
        selected = sorted(self.population, key=self.fitness, reverse=True)
        return selected[: self.population_size // 2]




    def crossover(self, parent1, parent2):
        point = random.randint(1, self.gene_length - 1)
        child = parent1[:point] + parent2[point:]
        return child




    def mutate(self, individual, mutation_rate=0.01):
        for i in range(self.gene_length):
            if random.random() < mutation_rate:
                individual[i] = random.random()
        return individual




    def evolve(self, generations):
        for _ in range(generations):
            selected = self.selection()
            new_population = selected[:]
            while len(new_population) < self.population_size:
                parent1 = random.choice(selected)
                parent2 = random.choice(selected)
                child = self.crossover(parent1, parent2)
                child = self.mutate(child)
                new_population.append(child)
            self.population = new_population
        best = max(self.population, key=self.fitness)
        return best, self.fitness(best)








class LSTM:
    def __init__(self, input_size, hidden_size, output_size):
        self.input_size = input_size
        self.hidden_size = hidden_size
        self.output_size = output_size
        self.W_i = [[random.random() for _ in range(input_size)] for _ in range(hidden_size)]
        self.U_i = [[random.random() for _ in range(hidden_size)] for _ in range(hidden_size)]
        self.b_i = [random.random() for _ in range(hidden_size)]
        self.W_f = [[random.random() for _ in range(input_size)] for _ in range(hidden_size)]
        self.U_f = [[random.random() for _ in range(hidden_size)] for _ in range(hidden_size)]
        self.b_f = [random.random() for _ in range(hidden_size)]
        self.W_o = [[random.random() for _ in range(input_size)] for _ in range(hidden_size)]
        self.U_o = [[random.random() for _ in range(hidden_size)] for _ in range(hidden_size)]
        self.b_o = [random.random() for _ in range(hidden_size)]
        self.W_c = [[random.random() for _ in range(input_size)] for _ in range(hidden_size)]
        self.U_c = [[random.random() for _ in range(hidden_size)] for _ in range(hidden_size)]
        self.b_c = [random.random() for _ in range(hidden_size)]
        self.W_y = [[random.random() for _ in range(hidden_size)] for _ in range(output_size)]
        self.b_y = [random.random() for _ in range(output_size)]




    def sigmoid(self, x):
        return 1 / (1 + math.exp(-x))




    def forward(self, inputs):
        h = [0] * self.hidden_size
        c = [0] * self.hidden_size




        i_gate = []
        for j in range(self.hidden_size):
            s = sum(inputs[k] * self.W_i[j][k] for k in range(self.input_size)) + \
                sum(h[k] * self.U_i[j][k] for k in range(self.hidden_size)) + self.b_i[j]
            i_gate.append(self.sigmoid(s))




        f_gate = []
        for j in range(self.hidden_size):
            s = sum(inputs[k] * self.W_f[j][k] for k in range(self.input_size)) + \
                sum(h[k] * self.U_f[j][k] for k in range(self.hidden_size)) + self.b_f[j]
            f_gate.append(self.sigmoid(s))




        o_gate = []
        for j in range(self.hidden_size):
            s = sum(inputs[k] * self.W_o[j][k] for k in range(self.input_size)) + \
                sum(h[k] * self.U_o[j][k] for k in range(self.hidden_size)) + self.b_o[j]
            o_gate.append(self.sigmoid(s))




        g_gate = []
        for j in range(self.hidden_size):
            s = sum(inputs[k] * self.W_c[j][k] for k in range(self.input_size)) + \
                sum(h[k] * self.U_c[j][k] for k in range(self.hidden_size)) + self.b_c[j]
            g_gate.append(math.tanh(s))




        c = [f_gate[j] * c[j] + i_gate[j] * g_gate[j] for j in range(self.hidden_size)]
        h = [o_gate[j] * math.tanh(c[j]) for j in range(self.hidden_size)]




        y = []
        for i in range(self.output_size):
            s = sum(h[j] * self.W_y[i][j] for j in range(self.hidden_size)) + self.b_y[i]
            y.append(self.sigmoid(s))
        return y








class Transformer:
    def __init__(self, d_model, num_tokens):
        self.d_model = d_model
        self.num_tokens = num_tokens
        self.W_q = [[random.random() for _ in range(d_model)] for _ in range(d_model)]
        self.W_k = [[random.random() for _ in range(d_model)] for _ in range(d_model)]
        self.W_v = [[random.random() for _ in range(d_model)] for _ in range(d_model)]
        self.W_o = [[random.random() for _ in range(d_model)] for _ in range(d_model)]




    def dot_product(self, a, b):
        return sum(x * y for x, y in zip(a, b))




    def matmul_vector(self, matrix, vector):
        return [sum(matrix[i][j] * vector[j] for j in range(len(vector))) for i in range(len(matrix))]




    def softmax(self, x):
        m = max(x)
        exps = [math.exp(i - m) for i in x]
        s = sum(exps)
        return [j / s for j in exps]




    def forward(self, inputs):
        queries = [self.matmul_vector(self.W_q, token) for token in inputs]
        keys = [self.matmul_vector(self.W_k, token) for token in inputs]
        values = [self.matmul_vector(self.W_v, token) for token in inputs]
        outputs = []
        for i in range(len(inputs)):
            scores = []
            for j in range(len(inputs)):
                score = self.dot_product(queries[i], keys[j]) / math.sqrt(self.d_model)
                scores.append(score)
            attn = self.softmax(scores)
            attn_output = [0] * self.d_model
            for j in range(len(inputs)):
                for k in range(self.d_model):
                    attn_output[k] += attn[j] * values[j][k]
            out = self.matmul_vector(self.W_o, attn_output)
            outputs.append(out)
        avg_output = [sum(x[k] for x in outputs) / len(outputs) for k in range(self.d_model)]
        proj_weights = [[random.random() for _ in range(self.d_model)] for _ in range(self.num_tokens)]
        proj_bias = [random.random() for _ in range(self.num_tokens)]
        token_scores = [
            sum(avg_output[k] * proj_weights[i][k] for k in range(self.d_model)) + proj_bias[i]
            for i in range(self.num_tokens)
        ]
        token_output = [1 / (1 + math.exp(-score)) for score in token_scores]
        return token_output








unique_words = list(set(words))
word_to_index = {word: i for i, word in enumerate(unique_words)}
index_to_word = {i: word for word, i in word_to_index.items()}




input_data = [[0] * len(unique_words) for _ in range(len(words) - 2)]
for i in range(len(words) - 2):
    input_data[i][word_to_index[words[i]]] = 1




output_data = [[0] * len(unique_words) for _ in range(len(words) - 2)]
for i in range(len(words) - 2):
    output_data[i][word_to_index[words[i + 1]]] = 1




input_size = len(unique_words)
hidden_size1 = round(PHI * input_size)
hidden_size2 = round(PHI * hidden_size1)
output_size = len(unique_words)




nn = NeuralNetwork(input_size, hidden_size1, hidden_size2, output_size)
epochs = round(100 * PHI)
for epoch in range(epochs):
    for i in range(len(input_data)):
        nn.forward(input_data[i])
        nn.backward(input_data[i], output_data[i], learning_rate=0.1)
    if (epoch + 1) % round(PHI) == 0:
        print("Feedforward NN Epoch {}/{}".format(epoch + 1, epochs))




rnn = RecurrentNeuralNetwork(input_size, hidden_size1, output_size)
rnn_output = rnn.forward(input_data[0])
print("Recurrent NN Output:", rnn_output)




kernel_size1 = round(3 * PHI)
kernel_size2 = round(2 * PHI)
cnn = ConvolutionalNeuralNetwork(input_length=round(10 * PHI), kernel_size1=kernel_size1,
                                 kernel_size2=kernel_size2, output_size=output_size)
sample_input = [random.random() for _ in range(round(10 * PHI))]
cnn_output = cnn.forward(sample_input)
print("Convolutional NN Output:", cnn_output)




population_size = round(10 * PHI)
ga = GeneticAlgorithm(population_size, round(PHI * 5))
best_individual, best_fitness = ga.evolve(round(50 * PHI))
print("Genetic Algorithm Best Individual:", best_individual, "Fitness:", best_fitness)




lstm_hidden_size = round(PHI * input_size)
lstm = LSTM(input_size, lstm_hidden_size, output_size)
lstm_output = lstm.forward(input_data[0])
print("LSTM Output:", lstm_output)




transformer_d_model = round(PHI * input_size)
transformer = Transformer(transformer_d_model, output_size)
transformer_input = []
for i in range(len(unique_words)):
    vec = [0] * transformer_d_model
    if i < transformer_d_model:
        vec[i] = 1
    transformer_input.append(vec)
transformer_output = transformer.forward(transformer_input)
print("Transformer Output:", transformer_output)








def advanced_text_generation(input_vector):
    ff_output = nn.forward(input_vector)
    rnn_out = rnn.forward(input_vector)
    lstm_out = lstm.forward(input_vector)
    transformer_out = transformer.forward([input_vector])
    combined = [
        (ff_output[i] + rnn_out[i] + lstm_out[i] + transformer_out[i]) / 4
        for i in range(len(ff_output))
    ]
    predicted_index = combined.index(max(combined))
    predicted_word = index_to_word[predicted_index]
    long_text = ""
    current_length = round(10 * PHI)
    for _ in range(5):
        segment = generate_text(current_length)
        long_text += segment + " "
        current_length = round(current_length * PHI)
    return long_text + predicted_word








def chat():
    print("FiPhi-NeuralMark ACC Initialized")
    base_length = round(5 * PHI)
    while True:
        user_input = input("\nYou: ")
        if user_input.lower() == "exit":
            print("Goodbye!")
            break
        user_input_tokens = user_input.split()
        input_vector = [0] * len(unique_words)
        for word in user_input_tokens:
            if word in word_to_index:
                input_vector[word_to_index[word]] = 1
        response = advanced_text_generation(input_vector)
        print("FiPhi-NeuralMark:", response)








chat()