Does nvComp support data_type for bfloat16 in python?

I would like to test compressing bfloat16 in Python. I found in the documentation that the device API supports this data type. However, I couldn't find any information in the documentation on how to compress bfloat16 using the Python API. Could you confirm whether this is supported?

import torch
import nvidia.nvcomp as nvcomp
import time

tensor = torch.rand(1000000, dtype=torch.bfloat16).cuda()

codec = nvcomp.Codec(algorithm="ANS", dtype="<f2")

start_time = time.perf_counter()
tensor_nvcomp = nvcomp.as_array(tensor)
compressed_tensor = codec.encode(tensor_nvcomp)
compression_time = time.perf_counter() - start_time

print(f"原始大小: {tensor.element_size() * tensor.numel()} bytes, 压缩后: {compressed_tensor.buffer_size} bytes")
print(f"压缩用时: {compression_time:.4f} 秒")

start_time = time.perf_counter()
decompressed_tensor_nvcomp = codec.decode(compressed_tensor, data_type="<f2")

decompressed_tensor = torch.as_tensor(decompressed_tensor_nvcomp, dtype=torch.bfloat16, device="cuda")

decompression_time = time.perf_counter() - start_time

print(f"解压用时: {decompression_time:.4f} 秒")

assert torch.allclose(tensor, decompressed_tensor), "false" # it return false to me

When I change torch.bfloat16 to float16, the assert passes successfully.

Additionally, where can I find a complete list of all supported decode data_type values? It took me a long time to discover that float16 corresponds to "<f2"...