Adds baseline implementation for MetaShuffling and some cleanups. (pytorch#4080)

Summary:

X-link: facebookresearch/FBGEMM#1164

Adds baseline implementation for MetaShuffling and some cleanups.

Differential Revision: D74101069

Author: levendlee

fbgemm_gpu/experimental/gen_ai/README.md

@@ -59,7 +59,7 @@ pip install fbgemm-gpu-genai
 
 ## 2.2 **Llama4 MoE support**
 
-More coming soon in [token shuffling](gen_ai/moe/README.md) kernels.
+More coming soon in [MetaShuffling](gen_ai/moe/README.md) kernels.
 
 # 3. **Llama 3 Related External Coverage**
 

fbgemm_gpu/experimental/gen_ai/gen_ai/moe/README.md

@@ -1,9 +1,15 @@
 # FBGEMM GenAI MoE Support
 
-MoE Token Shuffling Kernel support in FBGEMM GenAI Kernel Library.
+MetaShuffling MoE kernel support in FBGEMM GenAI kernel library.
 
-# **1. Overview**
+# **Overview**
 
-Mixture-of-Experts (MoE) is a popular model architecture for large language models (LLMs). Although it reduces computation in training and inference by activating less parameters per token,  it imposes additional challenges in achieving optimal computation efficiency with high memory and communication pressure, as well as the complexity to handle the dynamism and sparsity nature of the model. Here we introduce a new MoE inference solution, token shuffling, which enables us to efficiently deploy Llama 4 models for real scenario inference.
+Mixture-of-Experts (MoE) is a popular model architecture for large language models (LLMs). Although it reduces computation in training and inference by activating less parameters per token,  it imposes additional challenges in achieving optimal computation efficiency with high memory and communication pressure, as well as the complexity to handle the dynamism and sparsity nature of the model. Here we introduce a new MoE inference solution, MetaShuffling, which enables us to efficiently deploy Llama 4 models for real scenario inference.
 
 More technical design will be coming soon.
+
+# **Updates**
+
+- 2025-05-01: Initial version of MetaShuffling MoE pytorch example release.
+
+- 2025-04-17: Initial version of MetaShuffling MoE GPU kernels release.