77 4 months ago

qwen3 with YaRN RoPE scaling to 64k context (factor 2)

tools 30b
130163dbb64a · 136B
{
"num_ctx": 64000,
"repeat_penalty": 1,
"stop": [
"<|im_start|>",
"<|im_end|>"
],
"temperature": 0.6,
"top_k": 20,
"top_p": 0.95
}