Developed foundational Meta device support for int4 preshuffle kernels within the pytorch/FBGEMM repository, focusing on enabling PyTorch integration for quantized inference workloads. The work involved implementing meta kernels in C++—specifically preshuffle_i4_meta and f8i4bf16_shuffled_meta—to prepare and shuffle quantized data efficiently for Meta hardware. By establishing these data preparation and shuffling paths, the developer set the groundwork for end-to-end quantized inference and future performance benchmarking on Meta devices. This contribution broadened device compatibility for int4-quantized workloads in PyTorch, leveraging skills in C++, quantization, and Meta implementation to support device-specific optimizations and throughput improvements.