Add support for training per-LLM predictor

[James-QiuHaoran]

Support for training a customized predictor for a specific LLM model by adding a flag that specifies the model name from the dataset

[James-QiuHaoran]

Added in commits:

• https://github.com/James-QiuHaoran/LLM-serving-with-proxy-models/commit/21551d62e0172205045d256649dc103cca760ae8
• https://github.com/James-QiuHaoran/LLM-serving-with-proxy-models/commit/ae7a70a9d1d60f7e9cb13f8d317eb1d71d5054fc

[James-QiuHaoran]

Example commands:

# predictor training for a particular LLM model
python preprocess_dataset.py --task_type 0 --data_size 100 --model_name "gpt-4"
python latency_prediction.py --task_type 0 --data_size 100 --model_name "gpt-4"

Note that the model names are from LMSYS-Chat-1M dataset.