source/model.py

from __future__ import annotations

import math

import torch
from torch import nn


class ConditionalGenerator(nn.Module):
    def __init__(self, noise_dimensions: int = 32, embedding_dimensions: int = 16) -> None:
        super().__init__()
        self.noise_dimensions = noise_dimensions
        self.label_embedding = nn.Embedding(10, embedding_dimensions)
        self.network = nn.Sequential(
            nn.Linear(noise_dimensions + embedding_dimensions, 128),
            nn.LayerNorm(128),
            nn.SiLU(),
            nn.Linear(128, 128),
            nn.LayerNorm(128),
            nn.SiLU(),
            nn.Linear(128, 64),
            nn.Sigmoid(),
        )

    def forward(self, noise: torch.Tensor, labels: torch.Tensor) -> torch.Tensor:
        condition = self.label_embedding(labels)
        return self.network(torch.cat([noise, condition], dim=1))

    @torch.inference_mode()
    def generate(
        self,
        labels: torch.Tensor,
        *,
        seed: int,
        temperature: float = 1.0,
    ) -> torch.Tensor:
        generator = torch.Generator(device=labels.device).manual_seed(seed)
        noise = torch.randn(
            len(labels),
            self.noise_dimensions,
            generator=generator,
            device=labels.device,
        )
        return self(noise * temperature, labels)


class ProjectionCritic(nn.Module):
    def __init__(self, feature_dimensions: int = 64) -> None:
        super().__init__()
        self.features = nn.Sequential(
            nn.Linear(64, 128),
            nn.LeakyReLU(0.2),
            nn.Linear(128, feature_dimensions),
            nn.LeakyReLU(0.2),
        )
        self.score = nn.Linear(feature_dimensions, 1)
        self.label_projection = nn.Embedding(10, feature_dimensions)
        self.classifier = nn.Linear(feature_dimensions, 10)

    def forward(
        self,
        pixels: torch.Tensor,
        labels: torch.Tensor,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        features = self.features(pixels)
        projection = (features * self.label_projection(labels)).sum(dim=1)
        projection = projection / math.sqrt(features.shape[1])
        score = self.score(features).squeeze(1) + projection
        return score, self.classifier(features)


class TinyVisionJudge(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.features = nn.Sequential(
            nn.Conv2d(1, 8, kernel_size=3, padding=1),
            nn.GELU(),
            nn.Conv2d(8, 8, kernel_size=3, padding=1, groups=8),
            nn.GELU(),
            nn.Conv2d(8, 12, kernel_size=1),
            nn.GELU(),
            nn.MaxPool2d(2),
        )
        self.classifier = nn.Sequential(
            nn.Flatten(),
            nn.Linear(12 * 4 * 4, 10),
        )

    def forward(self, pixels: torch.Tensor) -> torch.Tensor:
        return self.classifier(self.features(pixels))


def parameter_count(model: nn.Module) -> int:
    return sum(parameter.numel() for parameter in model.parameters())