1 50 minutes ago

MeshStack routing assistant on sparse Qwen3-30B-A3B (MoE) with THOX Knowledge Pack. 53.75 tok/s long-form, 14.99 GB footprint on a GPU host at 8,192 ctx; grounding 8/8, trap declined. Config-only, not THOX-trained. Apache-2.0.

tools thinking
3af604ddd778 ยท 135B
{
"num_ctx": 8192,
"repeat_penalty": 1,
"stop": [
"<|im_start|>",
"<|im_end|>"
],
"temperature": 0.3,
"top_k": 20,
"top_p": 0.95
}