Oct 18 Today 14:47 Use BF16 state at every CUDA batch and stabilize physical-ring attention (#1 ) rockmatrix-ai +1: hkust-aaron - Use BF16 state at every CUDA batch and stabilize physical-ring attention (623a14eb5f8b3b9ac5ab9227975c6cb2ff0043ca)
Co-authored-by: Aaron Xu <hkust-aaron@users.noreply.huggingface.co> Oct 16 2 days ago 20:40 Fix Chinese README emphasis rendering 20:20 Use platform URLs for bilingual documentation navigation 20:15 Update report, runtime profiles and Chinese documentation 09:33 Remove legacy naming from report source 09:18 Align EdgeRock model card, figures and report with the current paper 04:11 Fix model-card navigation links 04:07 Use BF16 inference and align figures and results with the technical report Oct 9 9 days ago 18:23 Add portable Transformers inference and native vLLM and SGLang adapters 18:16 Use rockmatrix-ai Hugging Face model address 18:11 Align evaluation documentation and canonical release links RockMatrix
17:55 Publish complete 29-benchmark results with v43 scoring RockMatrix
Oct 8 10 days ago 07:05 Name evaluation resources explicitly RocoreMatrix
04:33 Keep cross-architecture comparisons benchmark-specific RocoreMatrix
03:58 Present full-split results across 27 common benchmarks RocoreMatrix
03:26 Release complete evaluation splits for all 29 text benchmarks RocoreMatrix
03:14 Clarify complete bundled evaluation defaults RocoreMatrix
02:50 Refine release header with centered title and icon navigation RocoreMatrix
02:44 Link technical report and add quickstart, deployment and evaluation guidance RocoreMatrix
Oct 7 11 days ago 17:37 Release YANchor-4B RocoreMatrix
