Use both fbgemm_gpu and fbgemm_gpu_genai

[wookjeHan]

Hi teams,
I need to use both torch.ops.fbgemm.jagged_to_padded_dense and torch.ops.fbgemm.quantize_fp8_per_row. To do this, I installed both fbgemm-gpu and fbgemm-gpu-genai using the following commands:

pip install fbgemm-gpu==1.3.0 --index-url https://download.pytorch.org/whl/cu128/
pip install fbgemm-gpu-genai==1.3.0 --index-url https://download.pytorch.org/whl/cu128/

However, it seems that I cannot use both operations. If I install only fbgemm-gpu, I can use torch.ops.fbgemm.jagged_to_padded_dense only, but if I install fbgemm-gpu-genai, I can only use torch.ops.fbgemm.quantize_fp8_per_row. And even when I install both fbgemm-gpu and fbgemm-gpu-genai, I cannot use both operations.

Is this behavior expected?

Thank you!

[q10]

Hi @wookjeHan this is expected behavior, and is a fallout from earlier refactoring work, where we made the the build system modular, i.e. build multiple targets. We will work on a fix for this.