Hardware and other requirements This artifact's performance evaluation requires an NVIDIA GPU. The core of the research involves running inference on Large Language Models (LLMs), which is not feasible for evaluation on commodity CPUs. Paper Replication Hardware (High-End): To replicate the exact performance metrics (e.g., ES, GS) for the largest models reported in the paper (Tables V, VI), hardware equivalent to the paper's testbed is needed: GPU: NVIDIA RTX 4090 (24GB VRAM) CPU: Intel Xeon Gold 6330 Recommended Minimum Hardware (for Llama 3 8B): To reproduce the main claims with the default Llama 3 8B model, a system with a modern NVIDIA GPU and at least 16GB of VRAM is recommended. Commodity Hardware / Functional Validation Path: For evaluators using commodity desktops or laptops with less powerful GPUs (e.g., 6GB-8GB VRAM), all claims can still be fully validated. The artifact scripts are configured to automatically use the gpt2 model (which is also evaluated in the paper, e.g., Table IV, VI) if the local path for Llama 3 8B is not found. gpt2will run on most modern commodity GPUs. Software Requirements OS: A modern Linux distribution (e.g., Ubuntu 22.04). Python: Version ≥3.10≥3.10. NVIDIA Stack: CUDA ≥11.8≥11.8 (as specified in requirements.txt). NVIDIA Driver compatible with the installed CUDA version. Python Packages: All dependencies are listed in requirements.txt. The main requirements are: torch transformers scipy pandas openpyxl Other Requirements Internet Connection: Required for installing Python packages via pip and for downloading the Hugging Face models. The gpt2 model (approx. 500MB) will be downloaded automatically.