Update README on benchmark script with correct AWS instances

scripts/benchmarks/README.md

@@ -13,8 +13,8 @@ For our paper, to achieve reproducible numbers, we have used the following AWS i
 | c6i.8xlarge | Intel | Ice Lake | Xeon 3 | 16 | 32 | 64GB | 300GB |
 | c6a.8xlarge | AMD   | Milan | Zen 3 | 16 | 32 | 64GB | 300GB |
 | c7i.8xlarge | Intel | Sapphire Rapids | Xeon 4 | 16 | 32 | 64GB | 300GB |
-| c7a.8xlarge | AMD   | Genoa with | Zen 4 | 16 | 16 | 32GB | 300GB |
-| c7g.8xlarge | Arm   | Graviton 3 | Neoverse V1 | 16 | 16 | 32GB | 300GB |
+| c7a.4xlarge | AMD   | Genoa with | Zen 4 | 16 | 16 | 32GB | 300GB |
+| c7g.4xlarge | Arm   | Graviton 3 | Neoverse V1 | 16 | 16 | 32GB | 300GB |
 
 All instances above use the Amazon Linux, which is free on AWS.
 
@@ -82,4 +82,4 @@ There are a number of things you can do to remove _"interim"_ state and re-start
  * If the `build_llvm.sh` script has been called, it will create a directory `~/installs/llvm` before everything. So, if the LLVM build fails, you have to remove it before running it again. You'll also have to remove the zip file (`<hash>.zip`) and the build directory (`llvm`) in the source directory of tpp-mlir.
  * If the LLVM build succeeds, then the install will be detected and you won't need to worry about it again.
  * If the TPP-MLIR build fails, you can troubleshoot by following the main README and build by hand. The script will always try to build again, but because we use `ninja`, this will take less than a second on an already built directory.
- * The benchmark scripts can show warnings or errors on its log. If that happens, go to the `benchmarks` directory and follow instructions on how to use the `driver.py` script and adjust for your system.
+ * The benchmark scripts can show warnings or errors on its log. If that happens, go to the `benchmarks` directory and follow instructions on how to use the `driver.py` script and adjust for your system.