GLM-OCR Local Deployment: Ollama vs vLLM vs SGLang

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull GLM-OCR model
ollama pull glm-ocr

# Interactive mode
ollama run glm-ocr

# Direct recognition
ollama run glm-ocr Text Recognition: ./document.png
ollama run glm-ocr Table Recognition: ./table.png

# Start Ollama server (runs by default)
ollama serve

# API call
curl http://localhost:11434/api/generate -d '{
  "model": "glm-ocr",
  "prompt": "Text Recognition:",
  "images": ["base64_encoded_image"]
}'

# Install vLLM
pip install -U vllm --extra-index-url https://wheels.vllm.ai/nightly

# Install Transformers (required for GLM-OCR)
pip install git+https://github.com/huggingface/transformers.git

vllm serve zai-org/GLM-OCR \
    --allowed-local-media-path / \
    --port 8080 \
    --tensor-parallel-size 1 \
    --max-model-len 8192

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8080/v1",
    api_key="not-needed"
)

response = client.chat.completions.create(
    model="zai-org/GLM-OCR",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": "file:///path/to/image.png"}},
            {"type": "text", "text": "Text Recognition:"}
        ]
    }],
    max_tokens=4096
)

pip install sglang
pip install git+https://github.com/huggingface/transformers.git

python -m sglang.launch_server \
    --model zai-org/GLM-OCR \
    --port 8080 \
    --tp 1

import sglang as sgl

@sgl.function
def ocr_recognition(s, image_path, task="Text Recognition:"):
    s += sgl.user(sgl.image(image_path) + task)
    s += sgl.assistant(sgl.gen("response", max_tokens=4096))

# Run
state = ocr_recognition.run(image_path="document.png")
print(state["response"])

FROM ollama/ollama

RUN ollama pull glm-ocr

EXPOSE 11434
CMD ["ollama", "serve"]

FROM vllm/vllm-openai:latest

ENV MODEL_NAME=zai-org/GLM-OCR

CMD ["--model", "${MODEL_NAME}", "--port", "8080"]

version: '3.8'
services:
  glm-ocr:
    image: vllm/vllm-openai:latest
    ports:
      - "8080:8080"
    volumes:
      - ./models:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    command: ["--model", "zai-org/GLM-OCR", "--port", "8080"]

import requests

def check_health(endpoint):
    try:
        response = requests.get(f"{endpoint}/health")
        return response.status_code == 200
    except:
        return False

upstream glm_ocr {
    server localhost:8080;
    server localhost:8081;
    server localhost:8082;
}

server {
    listen 80;
    location / {
        proxy_pass http://glm_ocr;
    }
}