llama.cpp qwen3.6

huihui.sh

#!/bin/bash

# Huihui Qwen 3.5 35B GGUF server for Apple M3 Max 36GB
# Optimized for lowest latency token generation

llama-server -hf cesarsal1nas/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M-GGUF:Q4_K_M \
  -c 65536 \
  -ngl 99 \
  -b 64 \
  -ub 1024 \
  -t 12 \
  --cache-type-k q8_0 \
  --cache-type-v q8_0

qwen_3.6.sh

#!/bin/bash

llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q4_K_M \
  -ngl 99 \
  -np 1 \
  -b 64 \
  -ub 1024 \
  -t 12 \
  --cache-type-k q8_0 \
  --cache-type-v q8_0 \
  --ctx-size 65536 \
  --temp 0.6 \
  --presence-penalty 0.0 \
  --top-p 0.8 \
  --top-k 20 \
  --reasoning off

qwen_3.6_mtp.sh

#!/bin/bash

llama-server -hf unsloth/Qwen3.6-35B-A3B-MTP-GGUF:UD-Q4_K_M \
  -ngl 99 \
  -np 1 \
  -b 128 \
  -ub 2048 \
  -t 12 \
  --cache-type-k q8_0 \
  --cache-type-v q8_0 \
  --ctx-size 65536 \
  --temp 0.6 \
  --presence-penalty 0.0 \
  --top-p 0.8 \
  --top-k 20 \
  --reasoning off \
  --spec-type draft-mtp \
  --spec-draft-p-min 0.9 \
  --spec-draft-n-max 2