Hi, I'm Abdullah 👋Full Stack Engineer & AI Developer

class=class="text-green">"text-faint italic"># Randomly mask class="text-orange">75% of patches, keep the rest
def random_masking(x, mask_ratio=class="text-orange">0.75):
    N, L, D = x.shape
    keep = int(L * (class="text-orange">1 - mask_ratio))
    idx = torch.argsort(torch.rand(N, L), dim=class="text-orange">1)
    return torch.gather(x, class="text-orange">1, idx[:, :keep, None].expand(-class="text-orange">1, -class="text-orange">1, D))

class=class="text-green">"text-faint italic"># ResNet50 encodes the image → LSTM decodes a caption
feats = resnet50(image)              class=class="text-green">"text-faint italic"># [B, class="text-orange">2048]
h = embed(feats).unsqueeze(class="text-orange">1)        class=class="text-green">"text-faint italic"># project to LSTM space
out, _ = lstm(word_embeddings, (h, c))
logits = fc(out)                     class=class="text-green">"text-faint italic"># next-word distribution

class=class="text-green">"text-faint italic"># Encrypt AES key with RSA public key
session_key = os.urandom(class="text-orange">32)         class=class="text-green">"text-faint italic"># AES-class="text-orange">256
cipher_rsa = PKCS1_OAEP.new(pub_key)
enc_key = cipher_rsa.encrypt(session_key)

class=class="text-green">"text-faint italic"># Forward noising step
def q_sample(x0, t, noise):
    a = alphas_cumprod[t]
    return a.sqrt() * x0 + (class="text-orange">1 - a).sqrt() * noise

struct Node {
    Node* next[class="text-orange">26] = {};
    bool end = false;
};
class=class="text-green">"text-faint italic">// insert: O(L), prefix search: O(L)