frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-21 05:34:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

graph mode: add hardswish inplace handling (#40284)

Browse Source 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40284

Adds graph mode handling for inplace hardswish, and test coverage for functional hardswish.

Test Plan:
```
python test/test_quantization.py TestQuantizeScriptPTSQOps.test_hardswish
```

Imported from OSS

Differential Revision: D22140628

fbshipit-source-id: 55a514f7dc1130d510f69ee4e611d7cb5e08d02e
This commit is contained in:
Vasiliy Kuznetsov
2020-06-21 09:35:44 -07:00
committed by Facebook GitHub Bot
parent c6dbfcaf9e
commit ab8a99bd36
4 changed files with 27 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
torch/csrc/jit/passes/quantization/quantization_patterns.h
View File

@ -848,6 +848,9 @@ graph(%a_quant, %alpha, %scale, %input_scale, %r_scale, %r_zero_point, %r_dtype)
auto hardswish = getObservedQParamOpFusionInfo(
"aten::hardswish", "quantized::hardswish", {}, {});
auto hardswish_ = getObservedQParamOpFusionInfo(
"aten::hardswish_", "quantized::hardswish", {}, {});
auto layer_norm = getObservedQParamOpFusionInfo(
"aten::layer_norm",
"quantized::layer_norm",
@ -968,6 +971,7 @@ graph(%a_quant, %alpha, %scale, %input_scale, %r_scale, %r_zero_point, %r_dtype)
{"quantized::mul", mul, quantized_mul},
{"quantized::mul", inplace_mul, quantized_mul},
hardswish,
hardswish_,
layer_norm,
group_norm,
instance_norm,