README.md

metadata

library_name: transformers
tags:
  - minicpm
  - intel
  - optimum-intel
  - tiny-random
  - openvino

Tiny-Random MiniCPM-o-2_6 for Testing

This is a minimal, randomly initialized version of the MiniCPM-o-2_6 architecture. It is designed specifically for CI/CD testing and unit tests in libraries like optimum-intel.

Model Details

• Architecture: MiniCPMForCausalLM
• Hidden Size: 24
• Layers: 1
• Precision: BF16
• Total Size: ~5.9 MB (under the 6MB requirement)

Intended Use

This model is NOT for actual inference or text generation. It is used to verify that integration pipelines (like OpenVINO conversion and quantization) are working correctly without the need to download several gigabytes of data.

Verification

The model has been verified to:

1. Load via AutoModelForCausalLM.
2. Perform a successful forward pass.
3. Fit within the strict 6MB size limit for lightweight testing suites.