Source files / install · raw

tersler.local.conf

# Local overrides for tersler.conf (see tersler.local.conf.example).

# Qwen3.8-27B Q6_K with the built-in MTP head (blk.64.nextn.*, checked 2026-11-06)
MODEL_URL="https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/Qwen3.8-27B-UD-Q6_K.gguf"
MODEL_SHA256="c9c206812fbe4ac7b76a729e25928b63f2ae89d37f69da7a71c20aec763cd436"
MODEL_HF_REPO="unsloth/Qwen3.8-27B-GGUF"
MMPROJ_URL="https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/mmproj-F16.gguf"
MMPROJ_FILE="Qwen3.8-27B-mmproj-F16.gguf"
MMPROJ_SHA256="cbb841a9ee0636b2ec172f5bb8df2ea8dfeb01e90fe7c6126581d662a0b4e43e"

# GPU intake fan (Arctic S12038-4K) on the NCT7904D: pwm2 -> fan5 (mapped 2026-11-06)
FAN_CONTROL=1
FAN_HWMON_NAME=nct7904
FAN_PWM=pwm2
FAN_INPUT=fan5

# serve the whole LAN (no API key: anyone on the LAN can use it)
LLAMA_HOST=0.0.0.0
# quiet curve: hotter card held at ~75 °C; full speed at 78 (slowdown is 82)
FAN_T_MIN=72
FAN_T_MAX=78

# Flash-Next on Strata beat the 27B on lean_bench (2026-11-10)
SERVE_ENGINE=strata