During July 2025, Xiaoruichao developed foundational Meta device support for int4 preshuffle kernels within the pytorch/FBGEMM repository. Leveraging C++ and PyTorch, Xiaoruichao implemented meta kernels—preshuffle_i4_meta and f8i4bf16_shuffled_meta—that prepare and shuffle quantized data for Meta hardware. This work established the data preparation and shuffling paths required for Meta-accelerated quantized inference, enabling PyTorch integration under the fbgemm namespace. By focusing on quantization and meta implementation, Xiaoruichao laid the groundwork for end-to-end quantized inference on Meta devices, addressing device compatibility and throughput for int4-quantized workloads while setting the stage for future performance optimizations.