# ThinkingCap NInfer V1 provenance

Current production artifact:

`combined-147590/thinkingcap-v1-upstream-d48dx3p5-oattndx6-ogdndx1-igdndx1-iattndx1-g64dx1-mtpgateupdx15p40-combined-147590.ninfer`

- Model SHA256: `d030303f73b658399810a1be097c3b682610c47df8ea1cd24eb0c1eb53255b27`
- Conversion sidecar SHA256: `62eff698b11f1d393b518cf62937f918dc131e6893f65ec9cb57d5c5d8a9843b`
- Published master: `ba8c4e726834c5e00cf5f68255ea6430c66e5319`
- Exact tested source tag: `thinkingcap-upstream-v1-2026-10-05`
- Exact tested source commit: `883d0e434ecb54cb6d6a9274c6a1a680988fab9d`
- Compact-vocab/Q8 commit: `b8685532`
- Runtime-DX commit: `883d0e43`
- CUDA architecture: `120a`
- Build type: `Release`
- Validated prefill chunk: `2048`
- Public vocabulary: `147590`
- Physical vocabulary: `147712`
- Proposal rows: `131072`
- Conversion device: CPU

The final baked artifact reproduced the runtime-tuned speculative acceptance
exactly at mixed 8k, 32k, 48k and 110k validation points.