Qwen3.8 27B
on 2× NVIDIA RTX 3090 · LM Studio · 16,384 ctx
Summary
User reports running GSM8K, MATH-500, HumanEval and MBPP locally.
User found reasoning stalls causing significant accuracy drops on hard math.
User notes the HumanEval score is possibly inflated due to contamination, and that harness bugs were fixed.