Unknown runtime specified nvidia (using CPU) about cosmos

[Rzfeng]

After fixing this error, another issue persists when building the model (still unsolved)

3c433cce2c59_cosmos_cosmos_1 | ===== BUILDING MODEL ======
3c433cce2c59_cosmos_cosmos_1 | Built backbone resnet50
3c433cce2c59_cosmos_cosmos_1 | Building downstream components via shape testing
3c433cce2c59_cosmos_cosmos_1 | done shape testing, building, attention mechanisms
3c433cce2c59_cosmos_cosmos_1 | built multi head attention
3c433cce2c59_cosmos_cosmos_1 | 8, 8, 2048
3c433cce2c59_cosmos_cosmos_1 | super called
3c433cce2c59_cosmos_cosmos_1 | going to build a 131072 by 1024 matrix of weights
3c433cce2c59_cosmos_cosmos_1 | built embeddings
3c433cce2c59_cosmos_cosmos_1 | done
3c433cce2c59_cosmos_cosmos_1 | Traceback (most recent call last):
3c433cce2c59_cosmos_cosmos_1 |   File "run.py", line 132, in <module>
3c433cce2c59_cosmos_cosmos_1 |     run_inference(f'{tmp}/images', f'{tmp}/cc_proposals', model_config, weights, xml, device)
3c433cce2c59_cosmos_cosmos_1 |   File "/app/infer/infer.py", line 23, in run_inference
3c433cce2c59_cosmos_cosmos_1 |     model.load_state_dict(torch.load(weights, map_location={"cuda:0": device_str}))
3c433cce2c59_cosmos_cosmos_1 |   File "/home/user/.local/lib/python3.6/site-packages/torch/serialization.py", line 358, in load
3c433cce2c59_cosmos_cosmos_1 |     return _load(f, map_location, pickle_module)
3c433cce2c59_cosmos_cosmos_1 |   File "/home/user/.local/lib/python3.6/site-packages/torch/serialization.py", line 532, in _load
3c433cce2c59_cosmos_cosmos_1 |     magic_number = pickle_module.load(f)
3c433cce2c59_cosmos_cosmos_1 | _pickle.UnpicklingError: invalid load key, 'v'.
3c433cce2c59_cosmos_cosmos_1 exited with code 1

[ankur-gos]

Stale, closing.