Here Qwen 3.8 27b zero shot the same prompt, and that with UD-Q4_K_XL quant not even full precision
It didn't work for you maybe because of the harness you used
This is cool! Qwen 3.8 on your Mac 🔥🔥🔥
You see, the secret behind Qwen 3.8 27B is all due to the GRPO distillation method, they use the outputs of Qwen 3.8 Max as a target for the RL,...
You see, the secret behind the Qwen 3.8 27B is all due to the GRPO distillation method, they use the Qwen 3.8 Max outputs as a target for the RL, and this makes the smaller model think as much as it needs to catch up with the larger one. 🧠
22.5 hours of iteration.
One local model. Qwen 3.8 27B ran through 90 × 15-minute iterations on a DGX Spark (Asus Ascent GX10). The result? A complete LEGO New York City skyline, built through relentless iteration.
You have reached the end of the archive
All of qwen38