README.md

---
license: gemma
base_model: gemma-4
library_name: transformers
pipeline_tag: text-generation
tags:
  - gemma
  - gemma-4
  - coder
  - mixture-of-experts
  - mxfp4
  - thinking
  - tool-use
  - long-context
---

# gemma-4-180b-a22b-coder

gemma-4-180b-a22b-coder is a Gemma 4 based coder model for software engineering, code
editing, Q/A, tool use, and long-context assistant workflows.

## model

- Family: `gemma-4`
- Variant: `coder`
- Model type: sparse Mixture-of-Experts language model
- Total logical text parameters: approximately `179.4B`
- Active logical text parameters per token: approximately `22.0B`
- Active experts per token: `108`
- Weight format: MXFP4 expert weights with BF16 shared weights
- Context: up to `200000` tokens in the listed vLLM configuration

## serving

```bash
CUDA_VISIBLE_DEVICES=0,1 vllm serve /path/to/gemma-4-180b-a22b-coder \
  --served-model-name vllm/doobee \
  --host 0.0.0.0 \
  --port 23333 \
  --dtype bfloat16 \
  --tensor-parallel-size 2 \
  --enable-expert-parallel \
  --max-model-len 200000 \
  --gpu-memory-utilization 0.96 \
  --trust-remote-code \
  --reasoning-parser gemma4 \
  --tool-call-parser gemma4 \
  --enable-auto-tool-choice \
  --default-chat-template-kwargs '{"enable_thinking": true}' \
  --generation-config vllm \
  --language-model-only \
  --skip-mm-profiling \
  --max-num-seqs 1 \
  --max-num-batched-tokens 8192
```

For clients that should not receive reasoning text, send
`"include_reasoning": false` in chat-completion requests.

## files

- `config.json`
- `generation_config.json`
- `tokenizer.json`
- `tokenizer_config.json`
- `chat_template.jinja`
- `model.safetensors.index.json`
- MXFP4/BF16 safetensor shards

## license

This model is released under the Gemma license.