What is CLIKA?
CLIKA offers a robust solution for optimizing AI models for deployment across diverse hardware platforms. Its Automatic Compression Engine (ACE) SDK functions like a universal compiler and optimizer, intelligently reducing model size and increasing speed without significant performance loss. This engine analyzes model architecture to apply customized optimizations, bypassing lengthy manual processes for faster deployment readiness.
Utilizing CLIKA results in substantial benefits, including significantly reduced memory footprints (up to 87%), drastically faster inference speeds (up to 12x), and considerable cost savings (up to 90%), all while maintaining model accuracy (≤ -1% loss). The platform supports a wide range of AI models, including Vision, Audio, and LLMs (under 15B parameters), and targets major hardware backends like Nvidia, Intel, AMD, and soon Qualcomm, ensuring broad compatibility through ONNX format support.
Features
- Automatic Compression Engine (ACE): Analyzes model architecture and applies customized optimizations automatically.
- Significant Size Reduction: Reduces AI model memory footprint by up to 87%.
- Speed Enhancement: Increases inference speed by up to 12x.
- Minimal Accuracy Loss: Preserves model performance with ≤ -1% accuracy drop post-compression.
- Broad Model Support: Compatible with Vision, Audio, LLMs (under 15B parameters), and custom models.
- Multi-Hardware Compatibility: Targets Nvidia (TRT, TRT-LLM), Intel & AMD (OpenVINO), Qualcomm (QNN, Genie coming soon), and other ONNX-compatible hardware.
- On-Premise Deployment Option: ACE SDK can run in on-premise or air-gapped environments for data privacy.
- Advanced Compression Techniques: Employs quantization, pruning, layer fusion, layer replacement, layer simplification, and redundancy removal.
Use Cases
- Deploying AI models efficiently on edge devices with limited resources.
- Optimizing large language models (LLMs) for faster response times.
- Reducing cloud computing costs for AI inference.
- Accelerating computer vision applications on various hardware.
- Improving the user experience of AI-powered applications through faster processing.
- Streamlining the MLOps pipeline by automating model optimization.
- Enabling AI deployment in privacy-sensitive, air-gapped environments.
FAQs
-
How does CLIKA compression works?
The Automatic Compression Engine (ACE) SDK functions like a universal compiler, optimizer, and translator for all AI models, targeting every major hardware backend. ACE automatically generates a unique compression plan for every model by analyzing its architecture alone, identifying and applying customized optimizations specific to that structure without needing background information on the model. -
What types of AI models does CLIKA's ACE support?
CLIKA supports all types of AI models, including custom and fine-tuned ones, with a current limitation of under 15B parameters. Support for larger models is planned. -
Would it work on my custom model?
Yes, the compression engine works on any AI model composed of supported layers. Refer to the documentation for the full list of supported layers. -
What if I can't share my model or data?
The ACE SDK can operate in on-premise or air-gapped environments, ensuring everything remains on your computers and CLIKA cannot access your private model or data. -
What types of hardware does CLIKA's ACE support?
Currently supported hardware includes Nvidia (TRT, TRT-LLM), Intel & AMD GPUs and CPUs (OpenVINO). Qualcomm (QNN, Genie) support is coming soon. CLIKA can support any hardware if its inference framework supports the ONNX format, by analyzing framework limitations and converting unsupported elements to optimized alternatives.
