Skip to content

[Relay, Quantization, TOPI] int8 dense on CUDA & Dense op quantization #2877

Merged
ZihengJiang merged 12 commits intoapache:masterfrom
vinx13:feature/quantize_dense
Apr 26, 2019
Merged

[Relay, Quantization, TOPI] int8 dense on CUDA & Dense op quantization #2877
ZihengJiang merged 12 commits intoapache:masterfrom
vinx13:feature/quantize_dense

Conversation

@vinx13
Copy link
Copy Markdown
Member

@vinx13 vinx13 commented Mar 22, 2019

  • Quantize dense layers in quantization pass
  • Add out_dtype to DenseAttrs to support mixed precision.
  • Add Int8 dense schedule on CUDA

@eqy @icemelon9 @ZihengJiang

@tqchen
Copy link
Copy Markdown
Member

tqchen commented Mar 27, 2019

cc @masahi @nishi-t @kazum @ajtulloch can you help review this PR?

@vinx13 vinx13 force-pushed the feature/quantize_dense branch 2 times, most recently from 190943a to 6a6082b Compare April 3, 2019 02:10
@vinx13
Copy link
Copy Markdown
Member Author

vinx13 commented Apr 9, 2019

@merrymercy @yzhliu could you also help review?

@vinx13 vinx13 force-pushed the feature/quantize_dense branch from 89f1154 to bbfa578 Compare April 11, 2019 06:03
Comment thread python/tvm/relay/quantize/quantize.py Outdated
@tqchen
Copy link
Copy Markdown
Member

tqchen commented Apr 18, 2019

@vinx13
Copy link
Copy Markdown
Member Author

vinx13 commented Apr 24, 2019

@ZihengJiang @tqchen can you also take a look?

Comment thread python/tvm/relay/quantize/quantize.py Outdated
@vinx13 vinx13 force-pushed the feature/quantize_dense branch from 7f7dbdf to ba0709e Compare April 25, 2019 23:54
@vinx13 vinx13 force-pushed the feature/quantize_dense branch from ba0709e to a4e26da Compare April 25, 2019 23:55
@ZihengJiang ZihengJiang merged commit cc09497 into apache:master Apr 26, 2019
@ZihengJiang
Copy link
Copy Markdown
Contributor

Merged, thanks for the hard-working

wweic pushed a commit to wweic/tvm that referenced this pull request May 13, 2019
apache#2877)

* Quantize dense layers

* Add out_dtype arggument to dense; Add dense_int8 on CUDA

* Add topi unittest of dense int8

* Fix relay

* Fix topi integration

* Fix quantization

* Update dense_rewrite

* Triger CI

* Change qconfig quantize_dense to quantize_op

* Fix

* Remove quantize_op from qconfig
wweic pushed a commit to neo-ai/tvm that referenced this pull request May 13, 2019
apache#2877)

* Quantize dense layers

* Add out_dtype arggument to dense; Add dense_int8 on CUDA

* Add topi unittest of dense int8

* Fix relay

* Fix topi integration

* Fix quantization

* Update dense_rewrite

* Triger CI

* Change qconfig quantize_dense to quantize_op

* Fix

* Remove quantize_op from qconfig
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants