b914816142ae2776f531be1c0b49812a0bfde91f,torchdiffeq/_impl/adams.py,,g_and_explicit_phi,#,36

Before Change

    device = curr_t.device

    g = torch.empty(k + 1, dtype=dtype, device=device)
    explicit_phi = []
    beta = torch.tensor(1, dtype=dtype, device=device)

    g[0] = 1
    c = torch.arange(1, k + 2, dtype=dtype, device=device).reciprocal()

After Change

    for j in range(1, k):
        beta = (next_t - prev_t[j - 1]) / (curr_t - prev_t[j]) * beta
        beat_cast = beta.to(implicit_phi[j][0])
        explicit_phi.append(tuple(iphi_ * beat_cast for iphi_ in implicit_phi[j]))

        c = c[:-1] - c[1:] if j == 1 else c[:-1] - c[1:] * dt / (next_t - prev_t[j - 1])
        g[j] = c[0]

Instances