|
Download PROVENANCE.md from cpchoi/modular-arithmetic-challenge: direct link, hf CLI and curl.
- Browser
- Download file 1.63 kB
-
https://huggingface.co/cpchoi/modular-arithmetic-challenge/resolve/main/PROVENANCE.md
- Command line
-
hf download hf://cpchoi/modular-arithmetic-challenge/PROVENANCE.md
-
curl -L -o PROVENANCE.md https://huggingface.co/cpchoi/modular-arithmetic-challenge/resolve/main/PROVENANCE.md
1.63 kB
| # Provenance | |
| - Submitted Model 1 commit: | |
| `c7cb870bfb68bc278be5fe81ec636c012f043227`. | |
| - Submitted Model 1 checkpoint SHA-256: | |
| `1cfd6f501fffd0092635221444a104d181cd5117a0c67c8c1ac484701bb7669a`. | |
| - V42 checkpoint SHA-256: | |
| `e3cca8d8499b9eeee29912fbf94af734a4e903bd3b49174babcdd095c829a021`. | |
| - Parameters: 412,204. | |
| - Tensor transfer: exact for reducer, MAC, subtraction, factor, cancellation | |
| judge/revisor, prime-two router, operation policy, and phase policy. | |
| - Scalar clock training: 1,200 Adam steps on all 513 initial-clock and all 513 | |
| decrement-clock labels; every distilled decision exactly matches Model 1. | |
| - Fresh/reload raw Tier 1--10: 300/300. | |
| - Sealed hard corpus SHA-256: | |
| `2f4453cdfe3bc1a96d811f8c1e17c1d58cac6a893d3a501d92cff5b6f4a30fbc`. | |
| - Sealed hard corpus: 960/960, including 911/911 nonzero answers. | |
| - Official CPU sandbox: 1,000/1,000 scored Tier 1--10 in 66.4 seconds. | |
| - Same-seed Model 1 controls: 69.0 seconds before and 93.3 seconds after. | |
| - Official result JSON SHA-256 for V42 and both controls: | |
| `a4f4e3de7e2d20d308b89d5c1e16b3304376d45f3b5227e3ccdb8f7a7773a10c`. | |
| - Exact-package randomized weights: 0/73 correct nonzero answers. | |
| - Official static checker, preprocessing isolation, determinism, and strict | |
| checkpoint loading: pass. | |
| The public evaluator image used for the comparison was | |
| `sha256:b026871750215787b4b2b3f0a1c7ca844d482ecb6bdd7de40ab1939568cbbf30` | |
| with 4 CPUs, 8 GB RAM, no network, read-only root filesystem, a 2 GB temporary | |
| filesystem, and a 300-second inference budget. All compared runs used master | |
| seed | |
| `3900c0de3900c0de3900c0de3900c0de3900c0de3900c0de3900c0de3900c0de`. | |