v1.2.0

Summary

This new version of Concrete ML adds support for hybrid deployment and K-nearest neighbor classification. Hybrid deployment with FHE is an approach that improves on-premise deployment by converting parts of the model to remote FHE computation, in order to protect model intellectual property (IP), ensure license compliance and facilitate usage monitoring. The 1.2 version also adds an improvement to the built-in neural networks, making them 10x faster out-of-the-box.

Links

Docker Image: zamafhe/concrete-ml:v1.2.0
pip: https://pypi.org/project/concrete-ml/1.2.0
Documentation: https://docs.zama.ai/concrete-ml

v1.2.0

Feature

• Enable import of fitted linear sklearn models (771c7ff)
• Support QAT models in hybrid model (526b000)
• Expose statuses to compile torch (8abddf6)
• Add KNN classifier in CML (1c33ec8)
• Add power of two scaling adapter for roundPBS (546fac9)
• Add hybrid FHE models (be6aa6e)

Fix

• Fix confusing print in CNN tutorial of advanced-examples (9136c47)
• Fix path parsing and default in hybrid serving (afd049a)
• Fix flaky padding test (6aaf5f0)
• Fix issues with OMP library (2b61846)
• Make sure structured pruning and unstructured pruning work well together (ada18ab)
• Fix structured pruning crash not caught by test (cafd8d1)
• Fix bad top1 accuracy in cifar_brevitas_training use case (f0a984e)
• Fix flaky double_fit test (3da6408)
• Remove workaround for simulating linear models (3f622bc)
• Re-compute quantization params when re-fitting linear models (3bad62e)

Documentation

• Fix and improve credit scoring use case example (e4db376)
• Update contribution part (f2822d1)
• Document KNN, PoT, Hybrid models (68a0b4c)
• Update mnist CNN (f80c90b)
• Update mnist Fully Connected example with PoT + rounding (6e3d003)
• Update cifar_brevitas_training accuracy using representative calibration set (39480ef)
• Correct n_bits markdown value in the LLM use case notebook (0cf1174)