Poor documentation for AIMET ONNX

quic / aimet

AIMET is a library that provides advanced quantization and compression techniques for trained neural network models.

https://quic.github.io/aimet-pages/index.html

Other

2.01k stars 363 forks source link

Open escorciav opened 2 months ago

escorciav commented 2 months ago

May I know how serious is the development of aimet_onnx? The documentation is extremely poor wrt to PyTorch 😢

Can we at least get proper & complete examples that anyone can run? Sorry in advance if it's is a page rendering error

Context I'm considering building tooling on top of aimet_onnx. Stakeholders aren't willing to share PyTorch &/Or Tensorflow code.

quic-mangal commented 2 months ago

@escorciav, sorry for this, for complete examples, you could refer to Example notebooks. In the meantime, we will work on fixing this.

escorciav commented 2 months ago

SG, Thanks for chiming in. Opened the issue & documenting my journey such that my future-self (or others) have pointers :)

escorciav commented 2 months ago

@quic-mangal may I know if QAT is supported off-the-shelf with onnx? The docs says "Link (no training)"

Again, I'm considering building tooling on top of aimet_onnx. Stakeholders aren't willing to share PyTorch &/Or Tensorflow code for various reasons.

quic-mangal commented 2 months ago

@escorciav QAT is not supported for ONNX.

escorciav commented 2 months ago

That's a bummer for my plans :sob: . Anyways, thanks for letting me know :raised_hands:

My assistant, chaggy, is "knowledgeable" till version 1.18. Chaggy often talks bs so I don't trust him :100: . But, he nailed it :wink:

BTW, do you provide the documentation as PDF? perhaps it's relevant for RAG & leveraging LLM for document Q&A

quic-mangal commented 2 months ago

We don't provide documentation as PDF ATM. But we will take this suggestion into account. Thanks.