The below table should be used as reference to understand which frameworks (and their respective versions) are currently supported in the Platform when targeting specific hardware choices.
If you have a use case with a specific hardware/framework requirement that is not supported, please communicate this to your Customer Success representative so our team is aware and works diligently for your use case needs!
Hardware | ONNX | TFLite | TVM | Tensorflow | TensorRT x Cuda (via ONNX-RT) on GPU | Pytorch |
Cloud: NVIDIA GPUs | Yes for all FP32 and INT8 models - (ONNX 1.8) | Not Available due to TFLite not supporting NVIDIA GPUs | Yes for all FP32 and INT8 models | Yes for FP32 and INT8 models uploaded in TF format - TF 2.0-2.6 | Yes for all FP32 and INT8 models - TensorRT 7.0 x CUDA 10.2 (A version upgrade to TensorRT 7.2 x CUDA 11.1 is coming soon) | Yes for all models converted to Torchscript |
Cloud: x86 | Yes for all FP32 and INT8 models (ONNX 1.8) | Yes for FP32 and INT8 models uploaded in TFLite format (TF 2.0-2.6) | Yes for all FP32 and INT8 models | Yes for FP32 and INT8 models uploaded in TF format - TF 2.0-2.6 | N/A | Yes for all models converted to Torchscript |