|
|
|
@@ -0,0 +1,27 @@ |
|
|
|
# Tuesday May 18, 2020 at 18:20pm GMT+8 |
|
|
|
|
|
|
|
## Agenda |
|
|
|
|
|
|
|
* Review the design for Sparse update for GatherV2. |
|
|
|
|
|
|
|
## Conference links |
|
|
|
|
|
|
|
https://meet.jit.si/MindSporeCimpilerSIG |
|
|
|
|
|
|
|
## Attendees |
|
|
|
|
|
|
|
* Pan Yifeng (Huawei) |
|
|
|
* Wei Luning (Huawei) |
|
|
|
* Zhao Zhiqiang (Huawei) |
|
|
|
* Huang Dongrun (Huawei) |
|
|
|
|
|
|
|
## Notes |
|
|
|
|
|
|
|
1. The long-term solution is to support the `SparseTensor` feature (inherited from `MetaTensor`), but the landing period will be longer. |
|
|
|
2. The quick support solution is to express this sparse structure with `DataClass`, and `Class` will be convert to `tuple` in `clean.cc`. The implement can reference `SequenceIterator`, and use `MultiType` to add a overload for add op and gradient update, if necessary. |
|
|
|
3. `GatherV2` can use the `ScatterNDUpdate` to update the gradient. |
|
|
|
|
|
|
|
## Action items |
|
|
|
|
|
|
|
* Make a demo with `DataClass`. |
|
|
|
* Design the full solution for `SparseTensor`. |