SmolLM3 Benchmark

Deployable Python script for benchmarking SmolLM3 on multilingual QA datasets (XQuAD, MLQA, GSM8K).
Evaluates models with Exact Match (EM), F1, inference time, memory usage, and a custom Resource Efficiency Index (REI).

---

✨ Features

• Benchmark Hugging Face models on multilingual QA datasets
• Supports reasoning modes: think vs no_think
• Tracks EM, F1, inference time, memory usage
• Computes REI (Resource Efficiency Index)
• Saves results to JSON/CSV for easy analysis
• Works on both CPU and GPU

---

📦 Requirements

Install dependencies with pip:

pip install transformers datasets torch psutil numpy tqdm