Commits · 957d5dee2b646868edff0bdefa6b54c44eeecc7d · gaoqiong / composable_kernel

11 Oct, 2023 2 commits
- resolve conflict · 957d5dee
  danyao12 authored Oct 11, 2023
  
  957d5dee
- G1/G2 -> G1Q/G1KV · 10836d41
  danyao12 authored Oct 11, 2023
  
  10836d41
10 Oct, 2023 6 commits
- Merge pull request #977 from ROCmSoftwarePlatform/mha-train-tiny-update · 3f4eae1d
  Dan Yao authored Oct 10, 2023
```
Some tiny updates
```
  3f4eae1d
- Align MakeArgumentPointer() interface of qloop_light_v1 · 9ac015bd
  Qianfeng Zhang authored Oct 10, 2023
  
  9ac015bd
- Revert the change with regard to defining dynamic buffer ygrad_gid_buf · 48a9c0b6
  Qianfeng Zhang authored Oct 10, 2023
  
  48a9c0b6
- More fix of LSEBasePtr for g_idx · 8c67fac1
  Qianfeng Zhang authored Oct 10, 2023
  
  8c67fac1
- Merge pull request #979 from ROCmSoftwarePlatform/mha-train-develop-remove-fwd-deterministic · bf6b491e
  Dan Yao authored Oct 10, 2023
```
flash attention remove fwd deterministic
```
  bf6b491e
- remove-fwd-deterministic · 454e9445
  letaoqin authored Oct 10, 2023
  
  454e9445
09 Oct, 2023 2 commits
- Fix the API pointer types in qloop_v1 device-op · 301a4d7a
  Qianfeng Zhang authored Oct 09, 2023
  
  301a4d7a
- adjust grouped kernels interface · 9574b34d
  danyao12 authored Oct 09, 2023
  
  9574b34d
08 Oct, 2023 2 commits
- Add debugging info in CheckValidity of qloop_v2 kernel · 4bfcd57f
  Qianfeng Zhang authored Oct 08, 2023
  
  4bfcd57f
- Some tiny updates · b41172a3
  Qianfeng Zhang authored Oct 08, 2023
  
  b41172a3
27 Sep, 2023 4 commits
- update 52 examples w/ mqa/gqa · 29398e70
  danyao12 authored Sep 27, 2023
  
  29398e70
- merge mha-train-develop · 617bdf3f
  danyao12 authored Sep 27, 2023
  
  617bdf3f
- split bwd mqa/gqa · 104aeabc
  danyao12 authored Sep 27, 2023
  
  104aeabc
- Merge pull request #943 from ROCmSoftwarePlatform/mha-train-develop-inference-forward · b23b3d71
  Dan Yao authored Sep 27, 2023
```
Add new flash attention forward for inference
```
  b23b3d71
26 Sep, 2023 13 commits
- change the check for vector load · 1af0a0a4
  letaoqin authored Sep 26, 2023
  
  1af0a0a4
- change 'fwd' inc file name to 'infer' · b19fe02b
  letaoqin authored Sep 26, 2023
  
  b19fe02b
- change mha infer class and file name · 6dbced07
  letaoqin authored Sep 26, 2023
  
  6dbced07
- update include file name · 63ea1d70
  letaoqin authored Sep 26, 2023
  
  63ea1d70
- rename the base class and file name · cef44211
  letaoqin authored Sep 26, 2023
  
  cef44211
- add check code for vectorload · d8998cbb
  letaoqin authored Sep 26, 2023
  
  d8998cbb
- add grouped example · 13a0c55d
  letaoqin authored Sep 26, 2023
  
  13a0c55d
- start adding grouped example and device op · 9ca20e2c
  letaoqin authored Sep 26, 2023
  
  9ca20e2c
- change file run_grouped_multihead_attention_bias_forward.inc name · 7b13018e
  letaoqin authored Sep 26, 2023
  
  7b13018e
- train mqa/gqa · 198558c5
  danyao12 authored Sep 26, 2023
  
  198558c5
- fwd mqa/gqa · 6a2d7c9f
  danyao12 authored Sep 26, 2023
  
  6a2d7c9f
- clean up codes · c459f488
  danyao12 authored Sep 26, 2023
  
  c459f488
- bwd mqa/gqa w/ permute · 174e013d
  danyao12 authored Sep 26, 2023
  
  174e013d
25 Sep, 2023 8 commits
- fix class name · c703632f
  letaoqin authored Sep 25, 2023
  
  c703632f
- rename example file name · 6bc73d41
  letaoqin authored Sep 25, 2023
  
  6bc73d41
- add bias 0 · d9d68abf
  letaoqin authored Sep 25, 2023
  
  d9d68abf
- Merge branch 'mha-train-develop' of... · b15eecba
  letaoqin authored Sep 25, 2023
```
Merge branch 'mha-train-develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into mha-train-develop
```
  b15eecba
- start adding mutiple head flash attention forward for inference · 87d1e073
  letaoqin authored Sep 25, 2023
  
  87d1e073
- bwd mqa/gqa w/o permute · 104f9da6
  danyao12 authored Sep 25, 2023
  
  104f9da6
- Merge pull request #941 from ROCmSoftwarePlatform/mha-train-tinyfix · 04c206da
  Dan Yao authored Sep 25, 2023
```
Two tiny fixes
```
  04c206da
- Two tiny fixes · d6d30a86
  Qianfeng Zhang authored Sep 25, 2023
  
  d6d30a86
22 Sep, 2023 2 commits
- Merge pull request #905 from ROCmSoftwarePlatform/mha-train-develop-grad-bias · c0c52268
  Dan Yao authored Sep 22, 2023
```
flash attention output   bias grad
```
  c0c52268
- change d0 operator variables name · c88d1173
  letaoqin authored Sep 22, 2023
  
  c88d1173
21 Sep, 2023 1 commit
- fix output error · 12c0f86a
  letaoqin authored Sep 21, 2023
  
  12c0f86a