Sep 30 16:32 Restore the original README style and keep release instructions concise 16:19 Use the minimal public inference package in the quickstart 16:07 Focus paper weights on ViT-L/B/S inference and preserve legacy downloads Sep 21 18:18 Keep paper ViT-B on model manifest v1 17:56 Record paper CUDA manifests and source comparison evidence 17:56 Add original paper ViT-B and ViT-S FP16 CUDA serving weights Sep 19 10:16 Add ChessQ demo and contact links to the public card Sep 15 12:59 Publish ViT-L FP16 manifest and synthetic CUDA reference 12:55 Release ViT-L FP16 CUDA weights and document platform deployment choices Sep 13 18:57 Publish Lite Core ML FP16 r2 tensor model ViT-S/14 at 644 px, float32 NCHW input and float32 [1,64,13] logits. SHA-256: e01695f8adb0de7216bb1aae3a20ac2bb2f988a6c269a430181440e5d6f3292f Sep 12 12:50 Restore original model card and add two ViT-S weight entries 12:48 Simplify model card to downloads, code link, citation and license 12:42 Add ViT-S/14 644 FP32 and INT8 ONNX weights Two self-contained ONNX models with input/output details and SHA-256 hashes in the model card. Inference code and setup instructions are hosted in the linked GitHub repository. Sep 1 08:16 Link arXiv paper and add citation Aug 31 21:42 Update README.md 16:31 Use the paper caption below Figure 1 16:29 Show paper hero figure on model card 16:29 Add paper hero figure to model card 16:22 Update README.md 13:41 Add model weights license 13:41 Document released model weights 13:41 Add safe inference weights Aug 27 20:53 Complete model card 20:44 Add release checkpoint 14:04 initial commit 
