Spaces:
Sleeping
Sleeping
File size: 17,839 Bytes
e332630 e26a254 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 1ff2df0 e332630 6aeec30 e332630 77b8172 e332630 77b8172 e332630 9a5a685 e332630 9a5a685 e332630 9a5a685 e332630 10d5b41 e332630 1ff2df0 e332630 10d5b41 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 | import gradio as gr
import random
from typing import List, Set, Tuple, Optional
from dataclasses import dataclass
import time
# ========== КОНФИГУРАЦИЯ ==========
@dataclass
class Config:
num_neurons: int = 1000
depth: int = 3
input_size: int = 256
output_size: int = 256
spike_threshold: int = 1000
input_signal_strength: int = 1000
signal_scale_shift: int = 8
explore_noise_scale: int = 64
sparsity: float = 0.05
neighbor_radius: int = 10
hebbian_step_threshold: int = 10
sigma_decay_shift: int = 8
dopamine_decay: int = 9
homeostasis_interval_steps: int = 100
prune_interval_steps: int = 200
min_neurons: int = 100
target_activity: int = 100
# ========== НЕЙРОНЫ ==========
class Neuron:
def __init__(self, idx: int, threshold: int = 1000, noise_scale: int = 0):
self.idx = idx
self.potential = 0
self.threshold = threshold
self.fired = False
self.refractory = 0
self.neuron_sigma = 0
self.noise_scale = noise_scale
self.noise_accum = (idx * 1103515245 + 12345) & 0x7FFFFFFF
self.outgoing_synapses = []
self.outgoing_neurons = []
self.total_fires = 0
self.neuron_type = "basic"
def tick(self, step_count: int) -> Tuple[bool, int]:
if self.refractory > 0:
self.refractory -= 1
return False, 0
if self.potential >= self.threshold:
self.fired = True
output = self.potential
self.refractory = 2
self.neuron_sigma += 1
self.total_fires += 1
return True, output
self.fired = False
return False, 0
def integrate(self, signal: int):
if self.refractory == 0:
self.potential += signal
def fire_ternary(self) -> int:
if self.fired:
if self.potential >= self.threshold + (self.threshold >> 2):
return 1
elif self.potential <= self.threshold - (self.threshold >> 2):
return -1
return 0
def propagate(self, shift: int) -> List[Tuple[int, int]]:
if not self.fired:
return []
ternary = self.fire_ternary()
if ternary == 0:
self.potential = 0
return []
updates = []
for tgt, syn in zip(self.outgoing_neurons, self.outgoing_synapses):
signal = ternary * syn.weight
if signal != 0:
updates.append((tgt, signal << shift))
self.potential = 0
return updates
def get_logit_with_noise(self) -> int:
self.noise_accum = (self.noise_accum * 1103515245 + 12345) & 0x7FFFFFFF
if self.noise_scale == 0:
return self.potential
noise = (self.noise_accum % (self.noise_scale * 2 + 1)) - self.noise_scale
return self.potential + noise
def is_protected(self) -> bool:
return False
class ExcitatoryNeuron(Neuron):
def __init__(self, idx: int, threshold: int = 1000):
super().__init__(idx, threshold)
self.neuron_type = "excitatory"
def fire_ternary(self) -> int:
return 1 if self.fired else 0
class InhibitoryNeuron(Neuron):
def __init__(self, idx: int, threshold: int = 1000):
super().__init__(idx, threshold)
self.neuron_type = "inhibitory"
def fire_ternary(self) -> int:
return -1 if self.fired else 0
class OscillatorNeuron(Neuron):
def __init__(self, idx: int, period_steps: int, threshold: int = 1000):
super().__init__(idx, threshold)
self.neuron_type = "oscillator"
self.period_steps = period_steps
self.step_counter = 0
def tick(self, step_count: int) -> Tuple[bool, int]:
self.step_counter += 1
if self.step_counter >= self.period_steps:
self.step_counter = 0
self.fired = True
return True, self.threshold
return False, 0
def is_protected(self) -> bool:
return True
class CategoryDetectorNeuron(Neuron):
def __init__(self, idx: int, category_bytes: List[int], threshold: int = 1000):
super().__init__(idx, threshold)
self.neuron_type = "category_detector"
self.category_bytes = set(category_bytes)
def integrate(self, signal: int, input_byte: Optional[int] = None):
if input_byte is not None and input_byte in self.category_bytes:
super().integrate(signal)
def is_protected(self) -> bool:
return True
# ========== СИНАПС ==========
class Synapse:
def __init__(self, source: int, target: int, weight: int = 0):
self.source = source
self.target = target
self.weight = weight
self.G = 0
self.sigma = 0
self.bcm_theta = 512
def hebbian_update(self, pre_fired: bool, post_fired: bool,
post_potential: int, dopamine_gain: int = 512):
if pre_fired and post_fired:
self.sigma += dopamine_gain
if abs(self.sigma) >= 10:
delta = 1 if self.sigma > 0 else -1
self.G = max(-1, min(1, self.G + delta))
self.sigma = 0
else:
self.sigma = (self.sigma * 255) >> 8
def homeostasis(self, target_activity: int = 100):
current = abs(self.weight) * 100
if current > target_activity:
scale = (target_activity << 8) // current
self.weight = (self.weight * scale) >> 8
# ========== ГРАФ ==========
class Graph:
def __init__(self, config: Config):
self.config = config
self._n = config.num_neurons
self.synapses: List[Synapse] = []
self._incoming_count = [0] * self._n
self._build()
def _build(self):
for i in range(self._n):
for j in range(max(0, i - self.config.neighbor_radius),
min(self._n, i + self.config.neighbor_radius + 1)):
if i != j and random.random() < self.config.sparsity:
self.synapses.append(Synapse(i, j))
self._update_incoming_counts()
def _update_incoming_counts(self):
self._incoming_count = [0] * self._n
for syn in self.synapses:
self._incoming_count[syn.target] += 1
def get_incoming_count(self, idx: int) -> int:
return self._incoming_count[idx] if idx < len(self._incoming_count) else 0
def get_outgoing_count(self, idx: int) -> int:
return sum(1 for s in self.synapses if s.source == idx)
def get_all_synapses(self) -> List[Synapse]:
return self.synapses
def remove_neuron(self, idx: int):
self.synapses = [s for s in self.synapses if s.source != idx and s.target != idx]
self._update_incoming_counts()
@property
def n(self) -> int:
return self._n
@property
def num_edges(self) -> int:
return len(self.synapses)
# ========== ПЛАСТИЧНОСТЬ ==========
class Plasticity:
def __init__(self, graph: Graph, config: Config):
self.graph = graph
self.config = config
def hebbian_update(self, active_neurons: Set[int], potentials: List[int],
dopamine_gain: int = 512):
for src in active_neurons:
for syn in self.graph.synapses:
if syn.source == src:
post_active = syn.target in active_neurons
post_pot = potentials[syn.target] if post_active else 0
syn.hebbian_update(True, post_active, post_pot, dopamine_gain)
def homeostasis_all(self):
for syn in self.graph.synapses:
syn.homeostasis(self.config.target_activity)
# ========== ЯДРО AURA ==========
class AuraCore:
def __init__(self, config: Config):
self.config = config
self.graph = Graph(config)
self.plasticity = Plasticity(self.graph, config)
self.neurons: List[Neuron] = []
self._init_neurons()
self._init_category_detectors()
self._init_oscillators()
self._build_caches()
self.input_neurons = list(range(min(config.input_size, len(self.neurons))))
self.output_neurons = list(range(min(config.output_size, len(self.neurons))))
self.step_count = 0
self.active_neurons: Set[int] = set()
self.dopamine_trace = 0
self.dopamine_gain = 512
self.homeostasis_osc = self.neurons[-2]
self.prune_osc = self.neurons[-1]
def _init_neurons(self):
n = self.config.num_neurons
n_exc = int(n * 0.70)
n_inh = int(n * 0.15)
idx = 0
for _ in range(n_exc):
self.neurons.append(ExcitatoryNeuron(idx, self.config.spike_threshold))
idx += 1
for _ in range(n_inh):
self.neurons.append(InhibitoryNeuron(idx, self.config.spike_threshold))
idx += 1
while idx < n:
self.neurons.append(Neuron(idx, self.config.spike_threshold))
idx += 1
def _init_category_detectors(self):
self.neurons.append(CategoryDetectorNeuron(
len(self.neurons), list(range(48, 58)), self.config.spike_threshold
))
self.neurons.append(CategoryDetectorNeuron(
len(self.neurons), list(range(65, 91)), self.config.spike_threshold
))
self.neurons.append(CategoryDetectorNeuron(
len(self.neurons), list(range(97, 123)), self.config.spike_threshold
))
self.neurons.append(CategoryDetectorNeuron(
len(self.neurons), [9, 10, 32], self.config.spike_threshold
))
def _init_oscillators(self):
self.neurons.append(OscillatorNeuron(
len(self.neurons), self.config.homeostasis_interval_steps, self.config.spike_threshold
))
self.neurons.append(OscillatorNeuron(
len(self.neurons), self.config.prune_interval_steps, self.config.spike_threshold
))
def _build_caches(self):
for n in self.neurons:
n.outgoing_synapses = []
n.outgoing_neurons = []
for syn in self.graph.synapses:
if syn.source < len(self.neurons):
self.neurons[syn.source].outgoing_synapses.append(syn)
self.neurons[syn.source].outgoing_neurons.append(syn.target)
def forward(self, input_byte: int, reward: int = 0, explore: bool = True) -> int:
self.dopamine_trace = (self.dopamine_trace * self.config.dopamine_decay + reward * 10) // 10
self.dopamine_trace = max(0, min(1024, self.dopamine_trace))
self.dopamine_gain = 512 + self.dopamine_trace
if self.homeostasis_osc.tick(self.step_count)[0]:
self.plasticity.homeostasis_all()
if self.prune_osc.tick(self.step_count)[0]:
self._prune_isolated()
input_idx = self.input_neurons[input_byte % len(self.input_neurons)]
input_neuron = self.neurons[input_idx]
if isinstance(input_neuron, CategoryDetectorNeuron):
input_neuron.integrate(self.config.input_signal_strength, input_byte=input_byte)
else:
input_neuron.integrate(self.config.input_signal_strength)
self.active_neurons.clear()
for _ in range(self.config.depth):
fired_this_step = []
for i, n in enumerate(self.neurons):
if n.tick(self.step_count)[0]:
fired_this_step.append(n)
self.active_neurons.add(i)
for src in fired_this_step:
updates = src.propagate(self.config.signal_scale_shift)
for tgt, signal in updates:
if tgt < len(self.neurons):
self.neurons[tgt].integrate(signal)
potentials = [n.potential for n in self.neurons]
self.plasticity.hebbian_update(self.active_neurons, potentials, self.dopamine_gain)
logits = [0] * self.config.output_size
for byte in range(self.config.output_size):
idx = self.output_neurons[byte]
if idx < len(self.neurons):
logits[byte] = self.neurons[idx].get_logit_with_noise() if explore else self.neurons[idx].potential
next_byte = max(range(self.config.output_size), key=lambda i: logits[i])
self.step_count += 1
return next_byte
def _prune_isolated(self):
if self.graph.n <= self.config.min_neurons:
return
dead = []
for i, n in enumerate(self.neurons):
if i < self.config.input_size or n.is_protected():
continue
if self.graph.get_incoming_count(i) == 0 and self.graph.get_outgoing_count(i) == 0:
dead.append(i)
if dead and self.graph.n - len(dead) >= self.config.min_neurons:
for idx in sorted(dead, reverse=True):
self.graph.remove_neuron(idx)
del self.neurons[idx]
self._build_caches()
# ========== ИНИЦИАЛИЗАЦИЯ МОДЕЛИ ==========
print("🚀 Инициализация AURA...")
config = Config()
config.num_neurons = 200
config.depth = 3
core = AuraCore(config)
print(f"✅ Модель готова: {len(core.neurons)} нейронов, {core.graph.num_edges} синапсов")
# ========== ФУНКЦИИ ДЛЯ GRADIO ==========
def generate_text(prompt: str, steps: int = 100, temperature: float = 1.0, reward: int = 0) -> str:
"""Генерация текста из промпта."""
if not prompt:
prompt = "A"
# Устанавливаем шум в зависимости от температуры
core.config.explore_noise_scale = int(64 * temperature)
start_time = time.time()
# Подаём промпт
for ch in prompt:
core.forward(ord(ch), reward=reward, explore=True)
# Генерируем
result = list(prompt)
current = ord(prompt[-1])
for _ in range(steps):
current = core.forward(current, reward=reward, explore=True)
if 32 <= current <= 126 or current in (9, 10, 13):
result.append(chr(current))
elif current == 0:
result.append(' ')
else:
result.append(chr(32 + (current % 95)))
elapsed = time.time() - start_time
stats = f"""
⏱️ Время: {elapsed:.2f} сек
📊 Шагов: {steps}
🔥 Активных нейронов: {len(core.active_neurons)}
💪 Дофамин: {core.dopamine_trace}
"""
return ''.join(result), stats
def reset_model():
"""Сброс модели."""
global core
core = AuraCore(config)
return "✅ Модель сброшена"
# ========== ИНТЕРФЕЙС GRADIO ==========
with gr.Blocks(title="AURA — Спайковая нейросеть") as demo:
gr.Markdown("""
# 🧠 AURA — Спайковая нейросеть с Hebbian-обучением
Минимальная версия ядра AURA. Модель обучается на лету через локальную пластичность.
""")
with gr.Row():
with gr.Column(scale=2):
prompt_input = gr.Textbox(
label="📝 Промпт",
placeholder="Введите текст для генерации...",
value="Hello",
lines=2
)
with gr.Row():
steps_slider = gr.Slider(
label="📏 Длина генерации",
minimum=10,
maximum=500,
value=100,
step=10
)
temp_slider = gr.Slider(
label="🌡️ Temperature (креативность)",
minimum=0.1,
maximum=2.0,
value=1.0,
step=0.1
)
reward_slider = gr.Slider(
label="🏆 Reward (награда за шаг)",
minimum=0,
maximum=100,
value=0,
step=10
)
generate_btn = gr.Button("🚀 Сгенерировать", variant="primary", size="lg")
reset_btn = gr.Button("🔄 Сбросить модель", variant="secondary")
with gr.Column(scale=3):
output_text = gr.Textbox(
label="✨ Сгенерированный текст",
lines=10,
max_lines=20
)
stats_text = gr.Textbox(
label="📊 Статистика",
lines=5
)
generate_btn.click(
fn=generate_text,
inputs=[prompt_input, steps_slider, temp_slider, reward_slider],
outputs=[output_text, stats_text]
)
reset_btn.click(
fn=reset_model,
inputs=[],
outputs=[output_text]
)
gr.Markdown("""
---
### О модели
- **Архитектура**: Спайковая нейросеть на разреженном графе
- **Обучение**: Hebbian-пластичность (локальные правила)
- **Нейронов**: 800 (70% excitatory, 15% inhibitory)
- **Категориальные детекторы**: цифры, буквы, пробелы
""")
if __name__ == "__main__":
demo.launch(
server_name="0.0.0.0",
server_port=7860,
theme=gr.themes.Soft()
) |