Skip to content

Commit 12189fe

Browse files
loadamsloadams
authored
Merge branch 'master' into fix_z3
2 parents 996f88f + 7af3a4b commit 12189fe

File tree

24 files changed

+869
-100
lines changed

24 files changed

+869
-100
lines changed

‎.github/workflows/amd-mi200.yml‎

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -48,8 +48,6 @@ jobs:
4848
- name: Install (ROCm) apex
4949
run: |
5050
git clone https://github.com/ROCmSoftwarePlatform/apex.git
51-
cd apex
52-
git checkout torch_2.1_higher
5351
CURRENT_VER=$(git rev-parse HEAD)
5452
INSTALLED_VER=$(cat /blob/amd-apex/.venv_installed_version)
5553
if [[ "$CURRENT_VER" != "$INSTALLED_VER" ]]; then

‎.github/workflows/cpu-torch-latest.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,7 @@ concurrency:
1919

2020
jobs:
2121
unit-tests:
22-
runs-on: ubuntu-22.04
22+
runs-on: ubuntu-24.04
2323

2424
steps:
2525
- uses: actions/checkout@v4

‎.github/workflows/no-torch.yml‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,7 @@ permissions:
1919

2020
jobs:
2121
unit-tests:
22-
runs-on: ubuntu-22.04
22+
runs-on: ubuntu-24.04
2323

2424
steps:
2525
- uses: actions/checkout@v4
@@ -30,6 +30,7 @@ jobs:
3030
- name: Python environment
3131
run: |
3232
pip uninstall torch --yes
33+
pip install setuptools
3334
pip list
3435
3536
- name: Build deepspeed

‎.github/workflows/nv-pre-compile-ops.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -21,7 +21,7 @@ concurrency:
2121

2222
jobs:
2323
unit-tests:
24-
runs-on: ubuntu-22.04
24+
runs-on: ubuntu-24.04
2525
container:
2626
image: deepspeed/gh-builder:ubuntu1804-py38-torch1131-cu116
2727

‎.github/workflows/release.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@ on:
77

88
jobs:
99
deploy:
10-
runs-on: ubuntu-22.04
10+
runs-on: ubuntu-24.04
1111
environment: release-env
1212

1313
steps:

‎COMMITTERS.md‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,9 @@
1+
# DeepSpeed TSC Committers #
2+
3+
| Name | GitHub ID | Affiliation
4+
|--- | ---- | --- |
5+
| Olatunji Ruwase | [tjruwase](https://github.com/tjruwase) | Microsoft |
6+
| Logan Adams | [loadams](https://github.com/loadams) | Microsoft |
7+
| Masahiro Tanaka | [tohtana](https://github.com/tohtana) | Microsoft |
8+
| Jeff Rasley | [jeffra](https://github.com/jeffra) | SnowFlake |
9+
| Minjia Zhang | [minjiazhang](https://github.com/minjiazhang) | UIUC |

‎csrc/adagrad/cpu_adagrad.cpp‎

Lines changed: 15 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -17,9 +17,9 @@ static std::unordered_map<int, std::shared_ptr<void>> s_optimizers;
1717

1818
// C++ interface
1919

20-
template <typename ds_params_percision_t, typename ds_state_precision_t>
21-
void Adagrad_Optimizer::Step_1(ds_params_percision_t* _params,
22-
ds_params_percision_t* grads,
20+
template <typename ds_params_precision_t, typename ds_state_precision_t>
21+
void Adagrad_Optimizer::Step_1(ds_params_precision_t* _params,
22+
ds_params_precision_t* grads,
2323
ds_state_precision_t* _exp_avg_sq,
2424
size_t _param_size)
2525
{
@@ -56,9 +56,9 @@ void Adagrad_Optimizer::Step_1(ds_params_percision_t* _params,
5656
}
5757
}
5858

59-
template <typename ds_params_percision_t, typename ds_state_precision_t>
60-
void Adagrad_Optimizer::Step_4(ds_params_percision_t* _params,
61-
ds_params_percision_t* grads,
59+
template <typename ds_params_precision_t, typename ds_state_precision_t>
60+
void Adagrad_Optimizer::Step_4(ds_params_precision_t* _params,
61+
ds_params_precision_t* grads,
6262
ds_state_precision_t* _exp_avg_sq,
6363
size_t _param_size)
6464
{
@@ -104,9 +104,9 @@ int create_adagrad_optimizer(int optimizer_id,
104104
return 0;
105105
}
106106

107-
template <typename ds_params_percision_t, typename ds_state_precision_t>
108-
void Adagrad_Optimizer::Step_8(ds_params_percision_t* _params,
109-
ds_params_percision_t* grads,
107+
template <typename ds_params_precision_t, typename ds_state_precision_t>
108+
void Adagrad_Optimizer::Step_8(ds_params_precision_t* _params,
109+
ds_params_precision_t* grads,
110110
ds_state_precision_t* _exp_avg_sq,
111111
size_t _param_size)
112112
{
@@ -121,15 +121,15 @@ void Adagrad_Optimizer::Step_8(ds_params_percision_t* _params,
121121
(_param_size - rounded_size));
122122
}
123123

124-
template <typename ds_params_percision_t, typename ds_state_precision_t>
124+
template <typename ds_params_precision_t, typename ds_state_precision_t>
125125
void step_invoker(std::shared_ptr<Adagrad_Optimizer> opt,
126126
void* _params,
127127
void* grads,
128128
void* _exp_avg_sq,
129129
size_t _param_size)
130130
{
131-
opt->Step_8((ds_params_percision_t*)(_params),
132-
(ds_params_percision_t*)(grads),
131+
opt->Step_8((ds_params_precision_t*)(_params),
132+
(ds_params_precision_t*)(grads),
133133
(ds_state_precision_t*)(_exp_avg_sq),
134134
_param_size);
135135
}
@@ -139,12 +139,12 @@ std::map<std::tuple<c10::ScalarType, c10::ScalarType>,
139139
invokers;
140140

141141
// Fill map with template functions for each type
142-
template <class ds_params_percision_t, class ds_state_precision_t>
142+
template <class ds_params_precision_t, class ds_state_precision_t>
143143
void create_invoker()
144144
{
145-
invokers[std::tuple(c10::CppTypeToScalarType<ds_params_percision_t>(),
145+
invokers[std::tuple(c10::CppTypeToScalarType<ds_params_precision_t>(),
146146
c10::CppTypeToScalarType<ds_state_precision_t>())] =
147-
step_invoker<ds_params_percision_t, ds_state_precision_t>;
147+
step_invoker<ds_params_precision_t, ds_state_precision_t>;
148148
}
149149
struct InvokerInitializer {
150150
InvokerInitializer()

‎csrc/adam/cpu_adam_impl.cpp‎

Lines changed: 15 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -18,9 +18,9 @@ static std::unordered_map<int, std::shared_ptr<void>> s_optimizers;
1818

1919
// C++ interface
2020

21-
template <typename ds_params_percision_t, typename ds_state_precision_t>
22-
void Adam_Optimizer::Step_1(ds_params_percision_t* _params,
23-
ds_params_percision_t* grads,
21+
template <typename ds_params_precision_t, typename ds_state_precision_t>
22+
void Adam_Optimizer::Step_1(ds_params_precision_t* _params,
23+
ds_params_precision_t* grads,
2424
ds_state_precision_t* _exp_avg,
2525
ds_state_precision_t* _exp_avg_sq,
2626
size_t _param_size)
@@ -67,9 +67,9 @@ void Adam_Optimizer::Step_1(ds_params_percision_t* _params,
6767
}
6868
}
6969

70-
template <typename ds_params_percision_t, typename ds_state_precision_t>
71-
void Adam_Optimizer::Step_4(ds_params_percision_t* _params,
72-
ds_params_percision_t* grads,
70+
template <typename ds_params_precision_t, typename ds_state_precision_t>
71+
void Adam_Optimizer::Step_4(ds_params_precision_t* _params,
72+
ds_params_precision_t* grads,
7373
ds_state_precision_t* _exp_avg,
7474
ds_state_precision_t* _exp_avg_sq,
7575
size_t _param_size)
@@ -126,9 +126,9 @@ int create_adam_optimizer(int optimizer_id,
126126
return 0;
127127
}
128128

129-
template <typename ds_params_percision_t, typename ds_state_precision_t>
130-
void Adam_Optimizer::Step_8(ds_params_percision_t* _params,
131-
ds_params_percision_t* grads,
129+
template <typename ds_params_precision_t, typename ds_state_precision_t>
130+
void Adam_Optimizer::Step_8(ds_params_precision_t* _params,
131+
ds_params_precision_t* grads,
132132
ds_state_precision_t* _exp_avg,
133133
ds_state_precision_t* _exp_avg_sq,
134134
size_t _param_size)
@@ -145,16 +145,16 @@ void Adam_Optimizer::Step_8(ds_params_percision_t* _params,
145145
(_param_size - rounded_size));
146146
}
147147

148-
template <typename ds_params_percision_t, typename ds_state_precision_t>
148+
template <typename ds_params_precision_t, typename ds_state_precision_t>
149149
void step_invoker(std::shared_ptr<Adam_Optimizer> opt,
150150
void* _params,
151151
void* grads,
152152
void* _exp_avg,
153153
void* _exp_avg_sq,
154154
size_t _param_size)
155155
{
156-
opt->Step_8((ds_params_percision_t*)(_params),
157-
(ds_params_percision_t*)(grads),
156+
opt->Step_8((ds_params_precision_t*)(_params),
157+
(ds_params_precision_t*)(grads),
158158
(ds_state_precision_t*)(_exp_avg),
159159
(ds_state_precision_t*)(_exp_avg_sq),
160160
_param_size);
@@ -165,12 +165,12 @@ std::map<std::tuple<c10::ScalarType, c10::ScalarType>,
165165
invokers;
166166

167167
// Fill map with template functions for each type
168-
template <class ds_params_percision_t, class ds_state_precision_t>
168+
template <class ds_params_precision_t, class ds_state_precision_t>
169169
void create_invoker()
170170
{
171-
invokers[std::tuple(c10::CppTypeToScalarType<ds_params_percision_t>(),
171+
invokers[std::tuple(c10::CppTypeToScalarType<ds_params_precision_t>(),
172172
c10::CppTypeToScalarType<ds_state_precision_t>())] =
173-
step_invoker<ds_params_percision_t, ds_state_precision_t>;
173+
step_invoker<ds_params_precision_t, ds_state_precision_t>;
174174
}
175175
struct InvokerInitializer {
176176
InvokerInitializer()

‎csrc/includes/cpu_adagrad.h‎

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -14,9 +14,9 @@
1414
#include "simd.h"
1515

1616
#define STEP(SPAN) \
17-
template <typename ds_params_percision_t, typename ds_state_precision_t> \
18-
void Step_##SPAN(ds_params_percision_t* _params, \
19-
ds_params_percision_t* grads, \
17+
template <typename ds_params_precision_t, typename ds_state_precision_t> \
18+
void Step_##SPAN(ds_params_precision_t* _params, \
19+
ds_params_precision_t* grads, \
2020
ds_state_precision_t* _exp_avg_sq, \
2121
size_t _param_size);
2222

@@ -28,10 +28,10 @@ class Adagrad_Optimizer {
2828
}
2929
~Adagrad_Optimizer() {}
3030
#if defined(__AVX512__) or defined(__AVX256__)
31-
template <int span, typename ds_params_percision_t, typename ds_state_precision_t>
31+
template <int span, typename ds_params_precision_t, typename ds_state_precision_t>
3232
void Step_AVX(size_t* rounded_size,
33-
ds_params_percision_t* _params,
34-
ds_params_percision_t* grads,
33+
ds_params_precision_t* _params,
34+
ds_params_precision_t* grads,
3535
ds_state_precision_t* _exp_avg_sq,
3636
size_t param_size);
3737
#endif
@@ -61,15 +61,15 @@ class Adagrad_Optimizer {
6161
};
6262

6363
#if defined(__AVX512__) or defined(__AVX256__)
64-
template <int span, typename ds_params_percision_t, typename ds_state_precision_t>
64+
template <int span, typename ds_params_precision_t, typename ds_state_precision_t>
6565
void Adagrad_Optimizer::Step_AVX(size_t* rounded_size,
66-
ds_params_percision_t* _params,
67-
ds_params_percision_t* grads,
66+
ds_params_precision_t* _params,
67+
ds_params_precision_t* grads,
6868
ds_state_precision_t* _exp_avg_sq,
6969
size_t _param_size)
7070
{
7171
#if !defined(__AVX512__)
72-
if (std::is_same_v<ds_params_percision_t, c10::BFloat16> ||
72+
if (std::is_same_v<ds_params_precision_t, c10::BFloat16> ||
7373
std::is_same_v<ds_state_precision_t, c10::BFloat16>) {
7474
return;
7575
}

‎csrc/includes/cpu_adam.h‎

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -14,9 +14,9 @@
1414
#include "simd.h"
1515

1616
#define STEP(SPAN) \
17-
template <typename ds_params_percision_t, typename ds_state_precision_t> \
18-
void Step_##SPAN(ds_params_percision_t* _params, \
19-
ds_params_percision_t* grads, \
17+
template <typename ds_params_precision_t, typename ds_state_precision_t> \
18+
void Step_##SPAN(ds_params_precision_t* _params, \
19+
ds_params_precision_t* grads, \
2020
ds_state_precision_t* _exp_avg, \
2121
ds_state_precision_t* _exp_avg_sq, \
2222
size_t _param_size);
@@ -43,10 +43,10 @@ class Adam_Optimizer {
4343
~Adam_Optimizer() {}
4444

4545
#if defined(__AVX512__) or defined(__AVX256__)
46-
template <int span, typename ds_params_percision_t, typename ds_state_precision_t>
46+
template <int span, typename ds_params_precision_t, typename ds_state_precision_t>
4747
void Step_AVX(size_t* rounded_size,
48-
ds_params_percision_t* _params,
49-
ds_params_percision_t* grads,
48+
ds_params_precision_t* _params,
49+
ds_params_precision_t* grads,
5050
ds_state_precision_t* _exp_avg,
5151
ds_state_precision_t* _exp_avg_sq,
5252
size_t param_size);
@@ -106,16 +106,16 @@ class Adam_Optimizer {
106106
};
107107

108108
#if defined(__AVX512__) or defined(__AVX256__)
109-
template <int span, typename ds_params_percision_t, typename ds_state_precision_t>
109+
template <int span, typename ds_params_precision_t, typename ds_state_precision_t>
110110
void Adam_Optimizer::Step_AVX(size_t* rounded_size,
111-
ds_params_percision_t* _params,
112-
ds_params_percision_t* grads,
111+
ds_params_precision_t* _params,
112+
ds_params_precision_t* grads,
113113
ds_state_precision_t* _exp_avg,
114114
ds_state_precision_t* _exp_avg_sq,
115115
size_t _param_size)
116116
{
117117
#if !defined(__AVX512__)
118-
if (std::is_same_v<ds_params_percision_t, c10::BFloat16> ||
118+
if (std::is_same_v<ds_params_precision_t, c10::BFloat16> ||
119119
std::is_same_v<ds_state_precision_t, c10::BFloat16>) {
120120
return;
121121
}

0 commit comments

Comments
 (0)