odooclaw-vision-mlx

MLX (Apple Silicon) conversion of odooclaw-vision — the OdooClaw vision model for document parsing and invoice OCR. Same model as the GGUF release (nicolasramos/odooclaw-vision), converted to MLX format for native Apple Silicon inference via mlx-vlm.

Model

• Base: zai-org/GLM-OCR (MIT)
• Architecture: GlmOcrForConditionalGeneration (GLM-OCR family)
• Format: MLX (bf16 LLM weights, quantized 4-bit)
• Size: ~1.2 GB
• Runtime: mlx-vlm ≥ 0.32

Usage (mlx-vlm)

from mlx_vlm import load, generate

model, processor = load("nicolasramos/odooclaw-vision-mlx")

output = generate(
    model,
    processor,
    prompt="Extract the invoice fields: supplier, date, total, taxes.",
    image="path/to/invoice.jpg",
    max_tokens=512,
)
print(output)

OdooClaw integration

The OdooClaw OCR pipeline (4-layer model-agnostic) accepts any OpenAI-compatible vision endpoint. On Apple Silicon, serve this model with oMLX and point vision_base_url to it; on Linux, use the GGUF release with llama.cpp. See the OdooClaw scripts/setup-local.sh installer — it auto-detects the platform and downloads the correct format.

License

MIT. The underlying GLM-OCR model is MIT-licensed by Zhipu AI.