Benchmark models using PyTorch and OpenVINO
Optimize and train foundation models using IBM's FMS
Measure execution times of BERT models using WebGPU and WASM
Download a TriplaneGaussian model checkpoint
Create demo spaces for models on Hugging Face
Display benchmark results
Convert Hugging Face models to OpenVINO format
Predict customer churn based on input details
Teach, test, evaluate language models with MTEB Arena
Evaluate LLM over-refusal rates with OR-Bench
Request model evaluation on COCO val 2017 dataset
Analyze model errors with interactive pages
Submit deepfake detection models for evaluation
OpenVINO Benchmark is a tool designed to benchmark models using PyTorch and OpenVINO. It allows users to compare the performance of models run through different frameworks, providing insights into speed, accuracy, and resource usage. This tool is particularly useful for optimizing model inference in production environments.
• Gapless PyTorch and OpenVINO Integration: Directly compare model performance between PyTorch and OpenVINO.
• Automated Model Conversion: Seamlessly convert PyTorch models to OpenVINO format for benchmarking.
• Comprehensive Performance Metrics: Measures inference speed, latency, throughput, and memory usage.
• Customizable Workloads: Allows users to define specific input shapes and batch sizes for accurate benchmarking.
• Cross-Architecture Support: Supports benchmarking on CPUs, GPUs, and other specialized hardware.
• Detailed Reporting: Generates clear and actionable reports for performance analysis.
What models are supported by OpenVINO Benchmark?
OpenVINO Benchmark supports models developed in PyTorch and compatible with OpenVINO. Models must be exported in a compatible format for benchmarking.
Can I use OpenVINO Benchmark on non-Intel hardware?
Yes, OpenVINO Benchmark supports benchmarking on various architectures, including non-Intel devices.
How do I interpret the benchmarking results?
Results are presented in a detailed report that compares metrics like inference speed, memory usage, and latency. This helps in identifying the most optimized framework for your use case.