NVIDIA / NVIDIA/Model-Optimizer

How to set the right quant cfg ?

Open
#1,723 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
3.8k
Forks
604
Avg merge
2d 8h
Merged PRs (30d)
142

Description

I have an ONNX, I have some doubts:

1, If I don't consider prediction accuracy, what is the fastest inference time and how can I obtain it ?

2, I found that, FP8 quantization is not much faster than INT8 quantization, sometimes even slower.

3, If I choose int8+fp16 as computational precision, use the default int8-max quant config setting is not get good performance, so do I still need to manual adjust the config based on the info of the engine by trex ?

4, Set autotune: bool = True in modelopt.onnx.quantization.quantize() get worse performance than manual setting ?

Thanks !

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with modelopt.onnx.quantization.quantize() and the int8-max configuration, then review the FP8, INT8, int8+fp16, autotune, and TREX cases described in the issue. Use the ONNX model and engine information referenced by the report to compare inference performance; done means documenting clear guidance for selecting or adjusting the quantization configuration.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Documentation
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.