This just made my week! I can't thank you enough for this amazing piece! Still somewhat speechless! 🙏❤ @XboxPope!

Final Fantasy VII Revelation, the thrilling conclusion of the FFVII Remake Series, launches simultaneously on PlayStation 5, Nintendo Switch 2, Xbox Series X|S, Epic Games Store and Steam in Spring 2027.

An optimized Qwopus3.6 27B v2 that consistently hits 30-40 tokens per second on STRIX HALO? Yes! CHADROCK QWOPUS3.6 27B v2 with MTP and @Italianclownz's ROCmFP4 optimizations is now the smartest and fastest 27B I've ever used. Highlights: HumanEval: • Chadrock: 159/164 = 96.95% • Original Q5_K_M: 151/164 = 92.07% HumanEval : • Chadrock: 155/164 = 94.51% • Original Q5_K_M: 147/164 = 89.63% That’s 8 tasks on HumanEval and 8 tasks on HumanEval vs the recorded original Qwopus3.6 27B v2 Q5_K_M row. Speed: • 164 HumanEval tasks • 45,033 completion tokens • 1346.8s cumulative request latency • ~59 tok/s mean total-token request speed • ~60 tok/s median total-token request speed • ~33.44 tok/s completion-only llama.cpp eval speed • ~37.14 tok/s peak active completion speed The original Q5_K_M run recorded 3834s generation time. Chadrock completed the same 164-task codegen workload with ~2.8x lower recorded request-generation time while also scoring higher. BFCL v4 non-live tool calling: • Overall: 85.88% • Simple Python AST: 94.50% • Multiple-call AST: 96.00% • Parallel-call AST: 86.50% • Parallel multiple-call AST: 85.50% • Irrelevance detection: 81.67% This is the profile to try if you want a local Strix Halo model that feels fast while staying sharp on coding and tool-use formats. You need Carlo's custom llama fork, this is built for AMD, details are in the model card: huggingface.co/jcbtc/qwopus3…