π“Ÿ host1.bitchan.one

𓀐 @*q?-8y6R

⏱ 4d seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

Recent Highlights from the --QFN failing bug-fix tasks and discussions on KV quantization: >110241970 >110242051 >110242069 >110242077 >110242091 >110242109 >110242152 >110242248 >110242314 >110242326 >110242334 >110242791 >110242857 >110242895 >110242905 --Release of delphi.cpp fork featuring hardware-aware NUMA support: >110240544 >110240576 >110240711 >110240768 >110240856 >110240917 --Repetition bugs caused by extreme iq2_xs quantization on Qwen: >110242880 >110242904 >110242913 >110242922 >110242928 >110242937 >110243288 >110243305 >110243321 >110243325 >110243115 --Analysis of model benchmarks and allegations of Kimi K3.1 contamination: >110241742 >110241748 >110241767 >110242137 >110241815 >110241932 --DeepSeek-V4.1-Flash Coder performance reports and benchmarks on RX 9060 XT: >110240908 >110240947 >110240975 >110240992 >110241048 --Hy4 graphics performance vs its restrictive safety filters: >110241041 >110241078 >110241241 >110241576 >110241595 >110241647 --Justifying $60k hardware spend to run Kimi k3.1 locally: >110243704 >110243725 >110243765 >110243732 >110243797 >110243849 --Widening performance gap between proprietary and open-source models: >110242635 >110242651 >110242706 >110242722 >110242764 >110242795 --Reasons for the stilted "code monkey" reasoning style in modern LLMs: >110239971 >110239982 >110239983 >110241608 >110240045 --MatterWave AI releases GRT-30M tiny Qwen distills: >110243939 >110244086 >110244107 --Amazon banning user for bulk Nvidia DGX Spark purchases: >110241100 >110241116 >110241123 >110241164 >110241197 --Logs: >110241672 >110242564 >110242994 >110243915 --Teto, Dipsy, Gemma (free space): >110240111 >110240207 >110240229 >110240288 >110240734 >110241671 >110241684 >110241884 >110241988 >110242089 >110242240 >110242449 >110242659 >110242911 >110243184 >110244049 Recent Highlight Posts from the Why?: Enable Links: https://rentry.org/lmg-recap-script⏱ 3d↩