A web tool that tells you whether your hardware can run a HuggingFace model before you download it.
Paste a model ID and get storage size, VRAM requirements across quantization formats, and a compatibility check against common GPUs.
Storage estimates for FP32, FP16, BF16, INT8, INT4, and 23 GGUF quantization variants (Q4_K_M, IQ3_M, Q8_0, and so on), based on parameter count and bits per weight.
VRAM requirements for each quantization format, with a separate KV cache estimate that updates as you adjust the context length slider (512 to 1M tokens).
Hardware compatibility across consumer GPUs (RTX 4090, 3090, 4080, 4070, 3080), prosumer cards (A4000, RTX 4000 Ada), cloud instances (A100, H100, A10G, L4, T4), Apple Silicon (unified memory), and CPU-only setups.
Private and gated models are supported. If a model requires authentication, the tool prompts for a HuggingFace access token. The token is held in memory for that session only and never stored anywhere.
Shareable links encode the model ID and selected quantization in the URL hash, so you can share a direct link to a specific result.
MIT