WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.
tree: /tank/aimodels/G4-MeroMero-v2-31B-heretic-NVFP4A16
tokenizer OK (1.6s)  truncation_side=right
Applying quantization config: 100%|██████████| 410/410 [00:00<00:00, 19976.82it/s]
Compressing model: 100%|██████████| 410/410 [00:01<00:00, 293.98it/s]
Loading weights: 100%|██████████| 2008/2008 [00:00<00:00, 4966.04it/s]
model loaded on CPU (2.7s)  Gemma4ForConditionalGeneration
parameters: 18.46 B
tensors still on meta device: 0  (none -- every weight was materialised)
prompt tokens: 23
Decompressing model: 100%|██████████| 410/410 [01:10<00:00,  5.85it/s]
generated 24 tokens in 82.1s (3.4s/tok, CPU)
OUTPUT: 'A lighthouse is a tower with a bright light used to guide ships at sea and warn them of dangerous coastlines.'