Inscribed by The Lonely Bit
Fractal Bitcoin Ordinals Qwen 2.5-7B q4f16 WebGPU

A full 7B inference stack, stored as ordinals in Fractal Bitcoin.

The first 7-billion-parameter model ever fully inscribed OnChain—and the one utilized in inAI v1. This atlas maps every ordinal that holds the public Qwen 2.5-7B-Instruct artifact compiled for browser WebGPU (MLC / WebLLM): weight shards, tokenizer, metadata, and the compiled compute kernel. Area = bytes on chain. Use the tree to browse like files; use the treemap for scale. Click any piece to open it on Fractal.

Flow: Ordinals → /content/<inscription id> → browser cache → WebGPU → tokens out
Treemap · bytes = area
01

Weights

88 params_shard_*.bin files hold the quantized parameters (~4 GiB). Large shards are split into many ordinal-sized pieces; the runtime reassembles them by byte order.

02

Tokenizer & metadata

Vocabulary, merges, and MLC chat config — small files, but required for the model to speak the same language as the base Qwen release.

03

WebGPU kernel

The compiled .wasm shader bundle tells the GPU how to run matmuls and attention for this exact graph. It is inscribed like any other asset.

Selected inscription