302 3 weeks ago

qwen3-coder with tools calling, context 58k to match full memory 31GB on RTX5090

tools
05a49f97461f · 162B
{
"min_p": 0,
"num_ctx": 58000,
"num_gpu": 999,
"repeat_penalty": 1.05,
"stop": [
"<|im_start|>",
"<|im_end|>"
],
"temperature": 0.3,
"top_k": 20,
"top_p": 0.8
}