Qwen3.6-27B

· llama.cpp · 8K context · Radeon PRO W7800 · 48 GB

Alibaba's dense Qwen3.6, 27B parameters.

53 tok/sdecode

– tok/sprefill

8Kcontext window

Earlier checkSep 6, 2026

Run it

Follow the steps below to download and run this recipe. Omarchy Local AI lists the recipes it can launch on your hardware. The server then answers on http://localhost:8000/v1.

1 Download the weights

hf download unsloth/Qwen3.6-27B-MTP-GGUF \
  --revision 5cb35eb3dcbf52dbce5f87dbc64df6aaffadcace

2 Write the server config

cat > qwen3.6-27b.llama.cpp.8k.yml <<'EOF'
{
  "llamacpp": {
    "backend": "rocm",
    "prefer_system": false,
    "rocm_args": "--flash-attn on -b 2048 -ub 512 -sm none --spec-type draft-mtp --spec-draft-n-max 3",
    "rocm_bin": "b1323"
  },
  "rocm_channel": "nightly"
}
EOF

3 Start the server

docker run --rm \
  --device /dev/kfd --device /dev/dri \
  -p 8000:13305 \
  --ipc host \
  --shm-size 16g \
  -v ~/.cache/huggingface:/root/.cache/huggingface \
  -v $PWD/qwen3.6-27b.llama.cpp.8k.yml:/opt/lemonade/.config/lemonade/config.json:ro \
  --entrypoint /opt/lemonade/lemond \
  ghcr.io/lemonade-sdk/lemonade-server:v11.9.0@sha256:7c780707cd695392a8680f557c7c1d55521383a71b1c4c5a9d9e6a226a1bee91 \
  --host 0.0.0.0

What it passed

✓ loadLoads and serves within an hour

✓ chatAnswers a plain question and stops on its own

– reasoningThinks separately and gets 17 × 23 right

– toolsCalls a tool with the right arguments and uses the result

– contextRecalls a code buried in a prompt that fills 85% of the window

– speedDecodes at 15 tok/s or more

Passed the older check (loads and chats); a full six-check run is pending.

Details

Weights: unsloth/Qwen3.6-27B-MTP-GGUF @ 5cb35eb3dc ›
Image: ghcr.io/lemonade-sdk/lemonade-server:v11.9.0@sha256:7c780707cd695392a8680f557c7c1d55521383a71b1c4c5a9d9e6a226a1bee91
Launch: llama.cpp-qwen3.6-27b-iq4-xs-8k ›
Recipe file: registry/recipes/amd/radeon-pro-w7800-48gb/qwen3.6-27b.llama.cpp.8k.json ›
Model released: Apr 21, 2026
Tested: earlier acceptance