15 days ago
-
Add config.json: model geometry, and the Hub query file that makes downloads countable
-
Link the published paper (arXiv:2608.23841); update citation
-
Align claim verdicts with project vocabulary: 6 Refuted, 8 Inconclusive (was Disproven - overclaimed a confounded test)
-
Remove non-public repo URL from citation
-
Fix critical-path bandwidth figure: naive is 9.00 MB/token (not 16.50); 9.00/4.50=2.00x
-
Fix model size: hosted model is arch2_4_8k_16l (~31B/16L), not 8.34B (that is the 4-layer scale point)
-
Upload model.cflow with huggingface_hub
-
Upload dist/windows-x86_64/cflow-convert.exe with huggingface_hub
-
Upload dist/windows-x86_64/cflow-run.exe with huggingface_hub
-
Upload dist/linux-x86_64/cflow-convert with huggingface_hub
-
Upload dist/linux-x86_64/cflow-run with huggingface_hub
-
Update README.md
-
initial commit
