File size: 1,400 Bytes
b19d528 | 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 | #include "sovereign_array.h"
#include <numeric>
namespace sovarr {
std::vector<size_t> unravel(size_t flat, const std::vector<size_t>& shape) {
std::vector<size_t> idx(shape.size());
size_t stride = 1;
for (size_t d = shape.size(); d-- > 0; ) {
idx[d] = (flat / stride) % shape[d];
stride *= shape[d];
}
return idx;
}
Array<float> softmax(const Array<float>& v) {
float s = 0.0f;
for (size_t i = 0; i < v.size(); ++i) s += std::exp(v[i]);
std::vector<float> out(v.size());
for (size_t i = 0; i < v.size(); ++i) out[i] = std::exp(v[i]) / s;
return Array<float>(v.shape(), std::move(out));
}
bool nand_gate(bool a, bool b) { return !(a && b); }
Array<float> nand_attention(const Array<float>& q, const Array<float>& k, const Array<float>& v) {
size_t n = q.shape()[0];
// scores_i = Σ_j q_i * k_j
std::vector<float> scores(n, 0.0f);
for (size_t i = 0; i < n; ++i)
for (size_t j = 0; j < n; ++j)
scores[i] += q[i] * k[j];
Array<float> scoresArr({n}, std::move(scores));
Array<float> w = softmax(scoresArr);
// out_i = Σ_j w_i * v_j
std::vector<float> out(n, 0.0f);
for (size_t i = 0; i < n; ++i)
for (size_t j = 0; j < n; ++j)
out[i] += w[i] * v[j];
return Array<float>({n}, std::move(out));
}
} // namespace sovarr
|