ModelVisio / Model formats
Model formats you can open in the browser
ModelVisio opens 39 neural-network model formats. 18 are fully parsed into an interactive graph with per-layer shapes, weights and FLOPs. The other 21 are detected and show the file's real metadata, never a made-up graph. Parsing runs in a Web Worker in your browser, so the model file is not uploaded.
Open a model or load the YOLO26n demo without a file.
Fully parsed: graph, inspector and pre-flight
| Format | File extensions |
|---|---|
| ONNX | .onnx |
| TFLite | .tflite .lite .tfl |
| Safetensors | .safetensors |
| GGUF | .gguf .ggml |
| NumPy | .npy .npz |
| Darknet | .cfg |
| PyTorch | .pt .pth .ckpt .bin |
| Core ML | .mlmodel .mlpackage |
| OpenVINO | .xml |
| TensorFlow | .pb .meta .pbtxt |
| Keras | .keras .h5 .hdf5 |
| Caffe | .caffemodel .prototxt |
| PaddlePaddle (.pdmodel) | .pdmodel |
| ncnn | .param |
| MNN | .mnn |
| RKNN | .rknn |
| MLIR | .mlir |
| scikit-learn | .pkl .joblib .pickle |
OpenVINO, ncnn, PaddlePaddle and TensorFlow.js keep their weights in a separate file. Select or drop the weights file together with the graph file (for example model.xml with model.bin).
Detected: metadata only
| Format | File extensions |
|---|---|
| ONNX Runtime | .ort |
| TorchScript | .ptl .torchscript |
| torch.export | .pt2 |
| ExecuTorch | .pte |
| Core ML (compiled) | .mlmodelc |
| TensorFlow.js | .json |
| MXNet | .params |
| PaddlePaddle params / Lite | .pdparams .nb |
| TNN | .tnnproto .tnnmodel |
| TensorRT | .engine .plan .trt |
| UFF | .uff |
| CNTK | .cntk .dnn |
| Barracuda | .nn |
| MegEngine | .mge .tm |
| NNabla | .nntxt |
| Hailo | .har .hn |
| Huawei CANN (OM) | .om |
| ML.NET | .mlnet |
| BigDL | .bigdl |
| CatBoost | .cbm |
| Darknet weights | .weights |
What you get for a fully parsed model
- A Netron-style graph with a per-layer inspector, exportable to SVG or PNG.
- A compiler pre-flight for Coral Edge TPU, RKNN, Hailo and Kneron (the rules).
- Hardware-fit scores for 21 edge accelerators: estimated FPS, FPS/W and memory fit.
- Exports to Graph JSON, Layers CSV, Safetensors and NumPy, plus TensorRT, HailoRT and RKNN deploy scripts.
Missing a format? Request it with a sample file.