OwLite
https://owlite.ai
Easy and Powerful AI Compression with OwLite
- OwLite incorporates SqueezeBits' unique compatibility technology between PyTorch, ONNX, and TensorRT.
- Converts PyTorch models to ONNX models and visualizes them to understand the model structure better.
- Converts ONNX model to Tensor RT and benchmarks it using GPUs registered with OwLite.
- This enables users to assess end-to-end latency and the latency contribution of each part within the visualized ONNX graph.
- Then, the user can compress the model based on the rich information mapped on the visualized ONNX.
- OwLite features powerful recommendations for beginners and fine-tuning even the most intricate details for advanced users.
- Furthermore, it provides an archive of compression results and analytical tools to boost the compression capabilities of individuals and organizations.
Start here!
- Start by following the 'Getting Started' section to set up OwLite.
- Then, dive into the 'Tutorial' section to guide you through your first project, showcasing OwLite's efficient project management capabilities.
- Get started now and experience the ease of OwLite!
OwLite python package
- The OwLite Python package can be seamlessly implemented into your existing PyTorch code.
- It generates a Compressed AI model based on the compression configuration edited on the web.
- With the compressed model, you can proceed through Post-Training Quantization (PTQ) and Quantization-Aware Training (QAT) processes.
- For detailed instructions on how to use this, please take a look at this link.
OwLite Web service
- The OwLite web editor visualizes the structure of your model, indicating the latency of specific parts of the model on your custom-selected device. It also informs you of the peak VRAM required to run the model.
- After gaining insights into the model,
- You can utilize SqueezeBits engineers' algorithms, optimized for the model's structure, to perform one-click automatic compression.
- If you possess a deeper understanding of AI models, OwLite also offers the capability to fine-tune down to the most detailed aspects according to your specific requirements.
- For detailed instructions on how to use this, please take a look at this link.
Contact
Please get in touch with owlite@squeezebits.com for any questions or suggestions.
AI model compression, AI model quantization, AI model optimization, AI compression, AI optimization, AI quantization