Does INT8 Quantized Convolution Still Involve Floating-Point Operations？
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 1.6k
- Forks
- 787
- PR merge metrics
- No merged PRs in 30d
Description
When I use quantized::conv2d in my model I noticed that a quantized convolution layer still keeps its scale parameter as a floating-point value. I think this scale is used to requantize the accumulated gemm output back to INT8. I would like to confirm:
Does the quantized convolution operator perform any floating-point computations internally, or is the entire operation carried out in pure INT8/INT32 arithmetic?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by tracing the quantized::conv2d implementation and the scale used for requantization. Determine whether the operator uses floating-point operations or only INT8/INT32 arithmetic, then document the confirmed behavior and relevant implementation details.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- machine-learning
- Issue type
- Documentation
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100