🧮 LLM Memory (VRAM) Estimator

Will this open model fit on my GPU, Mac or DGX Spark, and how fast can it possibly run? Pick a preset (architecture numbers come from each model's config.json) or enter your own.

Model preset Weight precision KV cache precision Context length (tokens) Batch / parallel sequences Overhead (%)

Architecture (auto-filled by preset)

Total parameters (billions) Active parameters per token (billions, MoE) Layers (num_hidden_layers) KV heads (num_key_value_heads) Head dim (head_dim or hidden_size / heads)

Hardware	Memory (GB)	Usable (GB, approx.)	Verdict	Speed ceiling