Why does GPU mode consume 160GB of memory but seem to be not using GPU at all?

[ggshuo647]

Hi, I installed the latest version from github on colab, and the CPU mode worked except a little bit slow. So I set the "use_gpu" to True and ran the code.
The hs.create_knn_graph was okay, but when I ran hs.compute_autocorrelations(), my RAM usage continued to rise and eventurally reached OOM (167G), while the VRAM remained stable around 0.4G. And I never saw the progress bar the whole time.
So I wonder whether I did something wrong?

My data is 700k cells and 13k genes, my code was like

hs = hotspot.Hotspot(
    adata_counts,
    model = 'danb',
    latent_obsm_key = "X_umap",
    use_gpu = True
)
hs.create_knn_graph(weighted_graph = False, n_neighbors = 30, approx_neighbors = True)
hs_results = hs.compute_autocorrelations()

Thanks in advance!