7b71ccb...
by
seongwoo
on 2022-09-23
[luci-interpreter] throw error when the memory limit exceeded (#9776)
This commit adds an exception of memroy limit exceeded.
ONE-DCO- 1.0-Signed- off-by: seongwoo <email address hidden>
3d20228...
by
chunseoklee <email address hidden>
on 2022-09-06
Release Note for 1.21.0 (#9674)
* Release Note for 1.21.0
- Release Note for 1.21.0
Signed-off-by: Chunseok Lee <email address hidden>
* Update docs/release/ 1.21/index. rst
Co-authored-by: seongwoo chae <email address hidden>
* Update docs/release/ 1.21/release- note_1. 21.0.md
Co-authored-by: Sanggyu Lee <email address hidden>
Signed-off-by: Chunseok Lee <email address hidden>
Co-authored-by: seongwoo chae <email address hidden>
Co-authored-by: Sanggyu Lee <email address hidden>
f59fd96...
by
chunseoklee <email address hidden>
on 2022-09-06
Update debian changelog to 1.21.0 (#9675)
- update debian changelog of compiler/runtime
Signed-off-by: Chunseok Lee <email address hidden>
Signed-off-by: Chunseok Lee <email address hidden>
8f632cd...
by
chunseoklee <email address hidden>
on 2022-09-06
Bulk layer's batch execution (#9668)
* Bulk layer's batch execution
Implement Bulk's batch execution. It supports parallel execution with 2
trix cores.
Signed-off-by: Chunseok Lee <email address hidden>
* Fix comment
* Strict assumption
* Use iteration in register Model
* Do not use constant core number
Signed-off-by: Chunseok Lee <email address hidden>
35905e1...
by
Sanggyu Lee <email address hidden>
on 2022-09-05
[onert] Add testcase dconv uint8 per-channel (#9663)
It adds testcase for dconv uint8 per-channel quantization.
ONE-DCO- 1.0-Signed- off-by: Hyeongseok Oh <email address hidden>
a5d18df...
by
chunseoklee <email address hidden>
on 2022-09-05
[onert] StaticShapeInferer for Bulk (#9666)
Note that Bulk's Static Shape Inferer assumes :
- new input shape has new batch size if it is different from model's
original input shape. Other dimension is not allowed to be changed.
- All new input shapes have the same batch. Thus, it is sufficient to
check 1st input's batch dim.
Signed-off-by: Chunseok Lee <email address hidden>
cdaad0f...
by
Sanggyu Lee <email address hidden>
on 2022-09-05
[onert] Support dconv per-channel uint8 (#9664)
It makes a way for dconv per-channel uint8 in operation layer.
ONE-DCO- 1.0-Signed- off-by: Sanggyu Lee <email address hidden>
d819f3f...
by
chunseoklee <email address hidden>
on 2022-09-05
Add field for Model's IO shapes in Bulk op (#9665)
- To enable shape inference of Bulk Operation, new fields are added in
Bulk op.
- TrixLoader loads new fields
Signed-off-by: Chunseok Lee <email address hidden>
df90b47...
by
Sanggyu Lee <email address hidden>
on 2022-09-02
[onert] Introduce dconv per-channel q8 uint kernel (#9662)
It introduces DepthwiseConvUI nt8.h from tflite/ reference/ integer_ ops
based on TensorFlow lite uint8 reference implementation.
It can handle per-channel quantized q8 weights.
ONE-DCO- 1.0-Signed- off-by: Sanggyu Lee <email address hidden>
9699963...
by
Hyeongseok Oh <email address hidden>
on 2022-09-02
[onert/test] Update GenModelTest fail message (#9661)
This commit changes GenModelTest fail message to use "expected" and "actual" for readability.
ONE-DCO- 1.0-Signed- off-by: Hyeongseok Oh <email address hidden>