!config.getFlag(BuilderFlag::kFP16). ) failure of TensorRT 10.6 when running demo_txt2img_flux.py on GPU h20

I want to run this demo in torch.float16, so I change utilities.py:
engine = engine_from_network(
network,
config=CreateConfig(fp16=True,
bf16=False,
tf32=False,
int8=False,
fp8=False,
refittable=enable_refit,
profiles=[p],
load_timing_cache=timing_cache,
**extra_build_args
),
save_timing_cache=timing_cache
)

error:
[E] IBuilder::buildSerializedNetwork: Error Code 3: API Usage Error (Parameter check failed, condition: !config.getFlag(BuilderFlag::kFP16). )
[!] Invalid Engine. Please ensure the engine was built correctly
polygraphy.exception.exception.PolygraphyException: Invalid Engine. Please ensure the engine was built correctly
/usr/lib/python3.10/tempfile.py:1008: ResourceWarning: Implicitly cleaning up <TemporaryDirectory '/tmp/tmpin7obfa_'>

Environment

cuda-python 12.6.0
cupy-cuda12x 13.2.0
dask-cuda 24.8.0
nvidia-dali-cuda120 1.41.0
pytorch-triton 3.0.0+dedb7bdf3
torch 2.5.0a0+b465a5843b.nv24.9
torch_tensorrt 2.5.0a0
torchprofile 0.0.4
torchvision 0.20.0a0

Start with demo_txt2img_flux.py and the engine_from_network configuration in utilities.py, then reproduce the TensorRT 10.6 build on the stated H20 and software environment. Trace why enabling torch.float16 produces the kFP16 API Usage Error; done means the demo builds a valid engine and runs successfully with the requested precision.