Local AI planning

Local AI Model Storage and Download Planner

Check model-library disk space and realistic transfer time before a download fills the drive or stalls a setup.

Saved only in this browser

When to use this

Before downloading a local model library

Use a published parameter count and intended quantization, then count similarly sized model files you expect to keep on disk.

Default result

The starter scenario represents two 14B files at 4-bit quantization with editable overhead, margin, disk, and bandwidth.

How the storage estimate works

One billion parameters at one bit is approximately 0.125 decimal gigabytes. The calculator multiplies parameter count by bits per parameter, then applies the file-overhead assumption and number of similarly sized files. Required free disk adds the selected safety margin and reserved capacity.

How to use the download estimate

The transfer calculation converts model-file gigabytes to megabits and divides by sustained Mbps. Treat it as an optimistic planning baseline because real downloads include protocol overhead, server limits, congestion, and interruptions.

FAQ

How is model size estimated?

The base estimate multiplies parameter count by quantization bits and converts bits to decimal gigabytes. File-format overhead is editable because metadata, tensors, and packaging vary.

Why add a storage safety margin?

Model managers may keep temporary downloads, multiple quantizations, caches, or an older copy during an update. The margin is planning headroom, not a filesystem requirement.

Is download time exact?

No. It assumes sustained bandwidth at the entered Mbps rate. Wi-Fi, server limits, protocol overhead, congestion, and interrupted downloads can make the real time longer.

Does this prove a model will run?

No. Disk capacity is separate from GPU VRAM, system RAM, context memory, and runtime compatibility. Continue with the Local AI Checker for memory fit.

Plan one model size

Before downloading a local model library

How the storage estimate works

How to use the download estimate

FAQ