lucloner/ qwen38-27b-turbofcfusion:mtp-q4km

625 2 weeks ago

DavidAU's uncensored multi-stage decensor of Qwen Team's Qwen3.8-27B: built-in MTP, native vision, 262K ctx (1M ext), thinking by default. Q4_K_M GGUF. Dual TITAN RTX @96.8K: 40 tok/s decode, acceptance 0.70. Cred: Qwen Team + DavidAU.

tools thinking
... /
params
d9ee97ba30c5 · 116B
{
"draft_num_predict": 5,
"num_batch": 2048,
"num_ctx": 262144,
"temperature": 0.6,
"top_k": 20,
"top_p": 0.95,
"use_mmap": true
}