Signal 3.8 27B AP-Q6_K GGUF
Observed on 2026-09-17, coding task text-only
Runtime: LM Studio bundled llama.cpp 2.37.0, commit 8172e65, standalone server; Hermes
Context: 118016
KV: q8_0 K/V
Native MTP: max 4; backend draft sampling disabled
Temperature: 0.6
Top-p: 0.95
Top-k: 20
Min-p: 0.05
Presence penalty: 1.0
BF16 vision projector loaded (not benchmarked for coding quality)
Result: 681.1753717500251 seconds; 91 tests, 0 failures; Debug build Pass
These are recorded settings, not a tested installation script. No isolated MTP speedup measured.