Repository navigation
Transformation from Nvidia to ggml #1693
the-code-samurai-97
started this conversation in
General
Replies: 1 comment
|
ggml doesn't include an ONNX importer, and GGUF wouldn't hold your network anyway: it stores tensors plus key/value metadata, and the GGUF spec lists a stored computation graph as a future extension, so the program that loads the file has to implement the architecture. In #762, slaren suggested starting from the PyTorch model rather than the ONNX export, since ONNX ops are lower level than most ggml ops. So the workflow becomes:
On current master, examples/mnist does exactly this: |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
My workflow is that use the python PyTorch to train the models and get the onnx file and use trtexec to convert to engine and load it with Nvidia cuda runtime and tensor\RT libraries.
How do I Change my workflow for ggml
and from Existing onnx file to gguf file is it possible
All reactions