Hello everyone! A few weeks ago, I released a 2.3bpw model intended for the 30xx 12 GB Ampere cards. Using a custom version of sjkvarn and several runtime changes, this made it possible to hit 200K ctx on a single 12GB card. Benchmarking placed it somewhere around qwen3.5 27b, but below 3.6. I've continued to work on quants for this size card and am happy to share that an updated checkpoint is now performing at Qwen3.6 BF16 levels on Livecode bench, with minimal post training (