huangxinjing
f17f3325c0
Fix cco and api comment
4 years ago
i-robot
dfd71f668b
!26556 modify parallel api note for master
Merge pull request !26556 from lilei/modify_parallel_API_note
4 years ago
lilei
f8827a09fb
modify parallel API note for master
4 years ago
i-robot
519f14a909
!26006 slice recompute activation
Merge pull request !26006 from yao_yf/add_transformer_slice_activation_config
4 years ago
huangxinjing
55efaabe2a
Fix api doc error
4 years ago
yao_yf
188d39da83
slice_activation_in_recompute
slice recompute activation
4 years ago
huangxinjing
e0674b37d3
Remove the comments
fix matmul name
4 years ago
huangxinjing
7bd77e549c
Fix number of the input params error
4 years ago
huangxinjing
4dc8e99d7c
Fix transformer dropout error
4 years ago
huangxinjing
ebdad2e450
Fix api comment
4 years ago
huangxinjing
a3ee85ede4
Fix reshape error
4 years ago
huangxinjing
827597f6fe
Add transformer example
4 years ago
huangxinjing
9ddd2b7669
Add past none check
Add if else for past none
4 years ago
linqingke
acde7febef
update pangu reshape and softmax performance.
Add layer norm judge
Fix layer norm name error
Fix input tyoe check
Fix ut test
Add 3d supports
4 years ago
huangxinjing
b787c5c8c8
Fix spell error
4 years ago
huangxinjing
31b3b46852
Replace TensorAdd with Add
4 years ago
i-robot
cdbe9b9a64
!23596 [Auto parallel] Move the MoE-related staff to an isolated file
Merge pull request !23596 from Xiaoda/89-moe-adaption
4 years ago
Xiaoda Zhang
615be06ec8
move moe-related staff to an isolated file
4 years ago
huangxinjing
0b89d5c9c4
fix batch size error
4 years ago
huangxinjing
c3a98bab2b
Add code check
fix pangu error
fix leakyrelu error
4 years ago
huangxinjing
7932c88aaf
fix message error
4 years ago
huangxinjing
110008e868
fix input error
4 years ago
huangxinjing
e02f553010
Fix spell error and add mode check
4 years ago
Xiaoda Zhang
5613c0b974
add a moe implementation:
1) extend the Liner cell for including BatchMatMul implementation, in
which the first dimension indicates the expert number;
2) implement a Switch (top1) router;
3) implement a MoE cell, which extends the FeedForward cell.
4 years ago
huangxinjing
6cea07f749
Add args check
4 years ago
huangxinjing
035eca2485
Fix performance degrades
4 years ago
huangxinjing
c7ec8c5f9f
Add value check and fix sparse brop
4 years ago
huangxinjing
75d16a4a97
hide doc comment of the construct
4 years ago
zhihenghu
ce12c02343
Add Sparse Attention
adjut the file structure and name
Deleted extra information
Do some formatting work
Add test case and fix some document
fix imports
4 years ago
huangxinjing
62496d75f3
less the interface exposed
4 years ago
huangxinjing
eaa8027903
Fix document api error
4 years ago
huangxinjing
d777742904
1. Move the class to mindspore.parallel, support activation sharding
4 years ago