update readme

Summary: as titled

Differential Revision: D82283600

Author: YifanYuan3

packages/fbgemm/README.md

@@ -1,9 +1,14 @@
-# FBGEMM Embedding Benchmark
+# FBGEMM Benchmarks
+
 `fbgemm_embedding` is a benchmark that simulates embedding table lookups representative of Model A and Model B workloads.
 The benchmark closely mimics the hotspot functions in table-based embedding (TBE) inference operations, which are commonly used in recommendation systems and deep learning models.
 
+`fbgemm_cpu_fp16` is a microbenchmark that evaluates the FP16-based matrix multiplication performance, which are common in Meta's recommendation and ranking related workloads.
+
+`fbgemm_cpu_spmdm8` is a C++ based micro-benchmark with some simple parameters and it benchmarks the TBE kernel. It's used for quick prototyping and performance validation.
+
 ## Installation
-To install the FBGEMM embedding benchmark, execute one of the following command:
+To install the FBGEMM benchmarks, execute one of the following command:
 ```bash
 ./benchpress -b ai install fbgemm_embedding_a_single
 ./benchpress -b ai install fbgemm_embedding_a_spec
@@ -14,7 +19,7 @@ To install the FBGEMM embedding benchmark, execute one of the following command:
 ./benchpress -b ai install fbgemm_cpu_spmdm8
 ```
 
-## Run FBGEMM Embedding Benchmark
+## Run FBGEMM Benchmarks
 ### Job - `fbgemm_embedding_a_single`
 `fbgemm_embedding_a_single` simulates the Model A workload using a single representative embedding table.
 This benchmark is designed to evaluate performance with a simplified embedding configuration.
@@ -29,14 +34,12 @@ This benchmark is designed to evaluate performance with a simplified embedding c
 `fbgemm_embedding_b_spec_int8` simulates the Model B workload using a representative set of embedding tables with int8 quantized precision.
 
 ### Job - `fbgemm_cpu_fp16`
-`fbgemm_cpu_a` executes the FP16Benchmark from FBGEMM-CPU.
+`fbgemm_cpu_a` executes the `FP16Benchmark` from FBGEMM.
 
 ### Job - `fbgemm_embedding_cpu_b`
-`fbgemm_cpu_b` executes the EmbeddingSpMDM8BitBenchmark from FBGEMM-CPU.
-This is a C++ based micro-benchmark which stresses the TBE kernel.
-This benchmark is used for quick prototyping and performance validation.
+`fbgemm_cpu_b` executes the `EmbeddingSpMDM8BitBenchmark` from FBGEMM.
 
-To run the FBGEMM embedding benchmarks, please use the following commands:
+To run the FBGEMM benchmarks, please use the following commands:
 ```bash
 ./benchpress -b ai run fbgemm_embedding_a_single
 ./benchpress -b ai run fbgemm_embedding_a_spec