one_click/README.md

# One-Click All-Local Deployment

This profile runs the full runtime without hosted model providers.

It starts:

- GLM-5.2 reasoning through local vLLM using `zai-org/GLM-5.2-FP8`.
- Qwen Omni vision/audio/video understanding through local vLLM-Omni using `Qwen/Qwen3-Omni-30B-A3B-Instruct`.
- Local OCR service built from this repo.
- Gateway API under `glm-5.2-visual-runtime`.
- PostgreSQL and MinIO for persistence.

## Start

```bash
cp one_click/.env.all-local.example one_click/.env.all-local
docker compose -f one_click/docker-compose.all-local.yml --env-file one_click/.env.all-local up --build
```

By default, vLLM downloads checkpoints from Hugging Face at startup. For a physically self-contained repo, first populate and upload `models/`, then set local paths in `one_click/.env.all-local`:

```bash
python scripts/materialize_weights.py --profile glm52_plus_qwen_omni
hf upload-large-folder wassemgtk/glm-5.2-visual-runtime models --type model
```

```bash
GLM_MODEL_PATH=/models/zai-org/GLM-5.2-FP8
QWEN_OMNI_MODEL_PATH=/models/Qwen/Qwen3-Omni-30B-A3B-Instruct
```

Gateway:

```text
http://localhost:8000/v1
```

Reasoning vLLM:

```text
http://localhost:8001/v1
```

Vision vLLM:

```text
http://localhost:8002/v1
```

OCR:

```text
http://localhost:8081/ocr
```

## Hardware

This is a real local deployment profile. GLM-5.2-FP8 and Qwen3-Omni are large checkpoints and require a serious multi-GPU environment. If you do not have enough GPU memory, the gateway can still run in `test` mode, but the all-local model services will not start.

## Smoke Test

```bash
python vllm/openai_smoke_test.py
```

For the gateway:

```python
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
response = client.responses.create(
    model="glm-5.2-visual-runtime",
    input=[{"role": "user", "content": [{"type": "input_text", "text": "Say ready"}]}],
)
print(response.output_text)
```