Skip to content
Snippets Groups Projects

Blocking user from using optimize sign while casting the graph/ Removing constant folding flag from ptq

Open Noam Zerah requested to merge noamzerah/aidge_quantization:Fix_ptq into dev
1 unresolved thread

Small MR introducing a few corrections:

  • Constant folding has been removed from the quantize_network() function and will soon be exposed independently via Pybind (Benjamin is currently working on this).

  • Added an assert to ensure that users do not enable the optimizeSign flag when casting the graph. This feature will be available in upcoming PTQ merge requests.

  • Updating the .gitignore file to exclude quantization_version.h

Edited by Noam Zerah

Merge request reports

Loading
Loading

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
1369 1368 }
1370 1369 if( targetType != DataType::Float64 && targetType != DataType::Float32 && targetType != DataType::Float16)
1371 1370 {
1372 AIDGE_ASSERT(!noQuant,"Cannot cast operators with the noQuant (Fake Quantization) flag set to true!")
1371 AIDGE_ASSERT(!noQuant,"Cannot cast operators with the noQuant (Fake Quantization) flag set to true!");
1372 AIDGE_ASSERT(!optimizeSigns,"Cannot cast operators with the optimize_sign flag set to true!")
  • changed milestone to %aidge v0.6.0

  • Noam Zerah added 1 commit

    added 1 commit

    Compare with previous version

  • changed milestone to %aidge v0.7.0

  • Please add a status label

  • Please register or sign in to reply
    Loading