Does executorch support qat quant for qualcomm qnn backend? #2169

Activity

1. cccclai

   added

   partner: qualcommFor backend delegation, kernels, demo, etc. from the 3rd-party partner, QualcommFor backend delegation, kernels, demo, etc. from the 3rd-party partner, Qualcomm

   on Mar 7, 2024
2. mergennachin

   added

   module: quantizationIssues related to quantizationIssues related to quantization

   partner: qualcommFor backend delegation, kernels, demo, etc. from the 3rd-party partner, QualcommFor backend delegation, kernels, demo, etc. from the 3rd-party partner, Qualcomm

   and removed

   partner: qualcommFor backend delegation, kernels, demo, etc. from the 3rd-party partner, QualcommFor backend delegation, kernels, demo, etc. from the 3rd-party partner, Qualcomm

   on May 13, 2024
3. jerryzh168 commented on May 16, 2024

   

   jerryzh168

   on May 16, 2024

   Contributor

   More actions

   QAT is going through the same code path as PTQ. if you don't have some special handling for conv-bn, you could start with adding the configuration for QAT, like: https://github.com/pytorch/pytorch/blob/a0429c01ad665ffb2faa04a411913ecee9962566/torch/ao/quantization/quantizer/xnnpack_quantizer.py#L116
4. jerryzh168 commented on Feb 4, 2025

   

   jerryzh168

   on Feb 4, 2025

   Contributor

   More actions

   @cccclai can you tag qualcomm POC to see if they want to support QAT?
5. cccclai commented on Feb 4, 2025

   

   cccclai

   on Feb 4, 2025

   Contributor

   More actions

   Yeah QAT is supported. cc: @shewu-quic @chunit-quic @haowhsu-quic @winskuo-quic let's add some docs

   👍React with 👍1Reacted by Jerry Zhang
6. pytorch

   locked and limited conversation to collaborators on Feb 5, 2025
7. byjlw

   converted this issue into a discussion #8215 on Feb 5, 2025