Tag
A user shares their experience running a quantized Qwen 3.8 27B model using QAT Q2 and Q5 KV, achieving high performance on a 12GB GPU with up to 200K token context, surpassing models like Sonnet 4.6.