fusorlabs/d1-omni-2b-web

d1-Omni-2B, compiled for the browser

Liquid AI's d1-Omni-2B compiled into a fusor-ai bundle: the weights the browser runtime of liquidai-lmf3-wasm loads to answer named, typed questions about a state, its images and its audio clips in one pass on the device's CPU. Try it in the Space fusorlabs/liquidai-lmf3-wasm.

• Source: LiquidAI/d1-Omni-2B at revision 4cccbfd5ba3405ee0f973134bd03e0d9d60ef1d8, compiled with the q7 profile: every large matrix 7-bit but the patch embedding and the audio subsampler's, kept in f32; with the checkpoint's calibration, without its speech output.
• Format: manifest.json lists every tensor and part; each part is named by its BLAKE3 hash, which the runtime checks as it loads.
• Reproduce it: just model in liquidai-lmf3-wasm downloads the checkpoint and compiles the same bundle.

This is not a PyTorch or safetensors checkpoint: it only loads in the liquidai-lmf3-wasm runtime.

The model is under the LFM Open License v1.0 (LICENSE).