wasi-nn
wasi-nn copied to clipboard
WebAssembly
Neural Network proposal for WASI
This merges together some of #69 with various discussions we've had in the [ML working group]. [ML working group]: https://github.com/bytecodealliance/meetings/blob/main/machine-learning/2024/ML-04-01.md
Is it possible to add an interface to get the tensor layout (memory format) of tensor? Such as `NCHW` or `NWHC`, I hope that incorrect arrangements can be determined and...
Any given model will have an embedded assumption about the version(s) of backends that will be able to consume it. There may be differences in one or more of: -...
The current tensor types are not all supported in core wasm. ``` enum tensor-type { FP16, FP32, FP64, BF16, U8, I32, I64 } ``` WASM supports the following core types:...
Hello! Firstly, thanks for the amazing work on enabling WebAssembly for ML [🤗](https://emojipedia.org/hugging-face)! I'm fairly new to adopting WebAssembly for machine learning, but I'm particularly interested to compile non-neural network...
Some models accept tensors whose items are `bytes`. In order to add these to `enum tensor-type`, we need to figure out how to represent these `bytes` items as tensor data,...
The initial proposal includes an [`execution_target`] flag to load the model for a specific type of device, e.g. `cpu`, `gpu`, `tpu`, etc. In the WASI CG meeting, several attendees discussed...
tensor-data is `list`. but i suppose that its actual contents are eg. f32. does wit/interface-types/etc provide a large enough alignment?
Which 16-bit float format are preferable for tensor's type? 1) IEEE half-precision 16-bit float (5 bits exponent, 10 bits fraction) 2) bfloat16 (8 bits exponent, 7 bits fraction), (supports by...
