BRIK conversion

Repacks a GGUF as BRIK — our web format (single file, aligned tensors, int8/int4 quants dequantized on the GPU). Lighter to download, faster to run, and above all instant loading: already quantized on disk, a BRIK goes straight to VRAM, skipping the conversion a GGUF redoes on every open.

Convert a model

Repacks a GGUF as BRIK (16-byte-aligned weights, sharded per layer, f16/int8/int4 tier of your choice). 100% in-browser conversion.

Choose a GGUF fileto convert to BRIK

Tokenizer / architecture:

Conversion profile:Recommended: near-identical quality to f16 at ~half the size (int8 dequantized on the fly).

Conversion keeps the whole model in f16 in memory — the tab may freeze for a while. Reserved for small/medium models.

Cached BRIK models

Automatic conversions (the "Convert to BRIK at load time" option in the app) are stored here for instant opens.

Loading…

Brimkern — open WebGPU engine, built by Romain Khanoyan. Local AI, WebGPU, on-device engines.