Gemma 4 12B

EXL3 4 bpw · tabbyapi · 128K context · RTX 4090 · 24 GB

Google's Gemma 4 12B from May 2026: a compact multimodal model with thinking, good with images and documents.

84 tok/sdecode

– tok/sprefill

128Kcontext window

Earlier checkSep 2, 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 turboderp/gemma-4-12B-it-exl3 \
  --revision e8d17ffbe696accdba72a47dac858d7e880d709e \
  --local-dir ~/models/gemma-4-12B-it-exl3-e8d17ffb

2 Write the server config

cat > gemma-4-12b.tabbyapi.128k.yml <<'EOF'
network:
  host: 0.0.0.0
  port: 5000
  disable_auth: true
  disable_fetch_requests: true
  send_tracebacks: false
  api_servers: ["OAI"]
  sse_ping_interval: 15

logging:
  log_prompt: false
  log_generation_params: false
  log_requests: false
  log_chat_completion_requests: false

model:
  model_dir: /workspace/models
  inline_model_loading: false
  use_dummy_models: false
  model_name: Gemma-4-12B-it-EXL3-4bpw
  backend: exllamav3
  max_seq_len: 131072
  cache_size: 131072
  cache_mode: Q6
  tensor_parallel: false
  gpu_split_auto: true
  autosplit_reserve: [192]
  chunk_size: 2048
  output_chunking: true
  max_batch_size: 2

draft_model:
  draft_mode: disabled

sampling:
  override_preset:

memory:
  sysmem_recurrent_cache: 4096
  sysmem_kv_cache: 0
  cuda_malloc_async: true
EOF

3 Start the server

docker run --rm \
  --gpus all \
  -p 8000:5000 \
  --shm-size 8g \
  -e NVIDIA_VISIBLE_DEVICES=all \
  -v ~/models/gemma-4-12B-it-exl3-e8d17ffb:/workspace/models:ro \
  -v $PWD/gemma-4-12b.tabbyapi.128k.yml:/app/config.yml:ro \
  --entrypoint /opt/venv/bin/python3 \
  ghcr.io/0xsero/tabbyapi-exl3@sha256:3d35e4979f5de7fd3b1621e4acd70035fa24ab78e941f4b569ab30fe1d1e7448 \
  main.py \
  --config /app/config.yml

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: turboderp/gemma-4-12B-it-exl3 @ e8d17ffbe6 ›
Image: ghcr.io/0xsero/tabbyapi-exl3@sha256:3d35e4979f5de7fd3b1621e4acd70035fa24ab78e941f4b569ab30fe1d1e7448
Launch: tabbyapi-gemma-4-12b-4-bpw-128k-v6 ›
Recipe file: registry/recipes/nvidia/rtx-4090-24gb/gemma-4-12b.tabbyapi.128k.json ›
Model released: May 23, 2026
Tested: earlier acceptance