A JavaScript tool that helps users determine if their GPU can run a specific LLM and what performance to expect. It calculates vRAM usage broken down by model size, KV cache, activation memory, and quantization overhead for GGML, bitsandbytes, and QLoRA methods. It also estimates token generation speed and fine-tuning iteration times, supporting inference frameworks like vLLM, llama.cpp, and HuggingFace.