Deterministic GGUF v2/v3 magic (0x46554747) KV-metadata & tensor-info parser + k-quant superblock byte-footprint calculator + Grouped-Query Attention (n_kv_heads) KV-cache & GDDR7/Unified-Memory roofline tok/s estimator engine (.gguf) built for llama.cpp, Ollama, LM Studio, NVIDIA GeForce — zero-upload ephemeral Wasm RAM execution.
Running ephemeral Wasm verification…