fervent_mcclintock/ qwen3.8-27b:iq3

20 yesterday

qwen3.8-27b:iq3 with a 260k context with a kv quant of q4_0 makes a pretty alright 16gb gpu coder in my 9070. It has managed fairly large projects with opencode and a custom goal loop plugin. Outputting faster than most affordable cloud models.

tools thinking
... /
params
6212f1f1ec2a · 178B
{
"num_ctx": 260000,
"num_predict": 2048,
"repeat_penalty": 1,
"stop": [
"<|im_start|>",
"<|im_end|>",
"<|im_end|>"
],
"temperature": 0.7,
"top_k": 20,
"top_p": 0.8
}