track RAM & save measurements during compile and run

[quic-morteza]

For the user to keep track of their measurements, code is added to log the necessary arguments and the measurements in addition to memory (RAM) usage during compile/runtime into a csv file. E.g. running the following command two times:

python -m QEfficient.cloud.infer --model_name gpt2 --batch_size 4 --prompt_len 64 --ctx_len 1024 --generation_len 512 --mxfp6 --num_cores 16 --device_group [0] --prompt "My name is|My name is|My name is|My name is" --benchmark

gpt2_benchmarking.csv is generated in the working directory that stores

[ochougul]

@quic-rishinr please review

[quic-rishinr]

Hi @quic-morteza could you please implement the benchmarking as two separate decorator functions? Each decorator should accept a custom_name: str (to specify stages like compilation, execution, etc.) and a benchmark: bool flag. One decorator should capture execution time, and the other should capture peak memory usage. This approach will help us scale without adding specific logic in each module for capturing stats.
Additionally please move the benchmark logging out of infer. The infer function should not contain any benchmarking-specific logic.

[quic-rishinr]

@quic-morteza any progress on the above request?