Why GPU memory isn't consumed?

I'm using 4090, have sage attn installed. It's still pretty fast, but this is weird

Unloaded DynamicSwap_LlamaModel as complete.
Unloaded CLIPTextModel as complete.
Unloaded SiglipVisionModel as complete.
Unloaded AutoencoderKLHunyuanVideo as complete.
Unloaded DynamicSwap_HunyuanVideoTransformer3DModelPacked as complete.
Loaded CLIPTextModel to cuda:0 as complete.
Unloaded CLIPTextModel as complete.
Loaded AutoencoderKLHunyuanVideo to cuda:0 as complete.
Unloaded AutoencoderKLHunyuanVideo as complete.
Loaded SiglipVisionModel to cuda:0 as complete.
section_index = 0, total_latent_sections = 100
Unloaded SiglipVisionModel as complete.
Moving DynamicSwap_HunyuanVideoTransformer3DModelPacked to cuda:0 with preserved memory: 6 GB
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [00:45<00:00, 1.81s/it]
Offloading DynamicSwap_HunyuanVideoTransformer3DModelPacked from cuda:0 to preserve memory: 8 GB
Loaded AutoencoderKLHunyuanVideo to cuda:0 as complete.
Unloaded AutoencoderKLHunyuanVideo as complete.
Decoded. Current latent shape torch.Size([1, 16, 10, 104, 60]); pixel shape torch.Size([1, 3, 37, 832, 480])
section_index = 1, total_latent_sections = 100

Start with the 4090 configuration and the loading, unloading, preserved-memory, and offloading messages in the report. Reproduce the run and inspect the reported GPU allocation around each model transition. Done means explaining whether the observed memory usage is expected and identifying the cause if it is not.