kmfoda commited on
Commit
0751425
·
verified ·
1 Parent(s): 2722729

Run 4. Outer Step 61. Inner Step 0. Peers 97.

Browse files
Files changed (4) hide show
  1. config.json +49 -49
  2. inner_optimizer.pt +1 -1
  3. model.safetensors +1 -1
  4. outer_optimizer.pt +1 -1
config.json CHANGED
@@ -1,23 +1,23 @@
1
  {
2
- "_name_or_path": "distributed/optimized-gpt2-1b",
3
  "activation_function": "gelu_new",
4
  "all_reduce_scores": {
5
  "0": "NON_PARTICIPATING",
6
  "1": "SUCCESS",
7
  "10": "NON_PARTICIPATING",
8
  "100": "NON_PARTICIPATING",
9
- "101": "NON_PARTICIPATING",
10
- "102": "NON_PARTICIPATING",
11
  "103": "NON_PARTICIPATING",
12
  "104": "SUCCESS",
13
- "105": "NON_PARTICIPATING",
14
  "106": "NON_PARTICIPATING",
15
- "107": "SUCCESS",
16
  "108": "NON_PARTICIPATING",
17
  "109": "NON_PARTICIPATING",
18
- "11": "NON_PARTICIPATING",
19
- "110": "NON_PARTICIPATING",
20
- "111": "SUCCESS",
21
  "112": "NON_PARTICIPATING",
22
  "113": "NON_PARTICIPATING",
23
  "114": "NON_PARTICIPATING",
@@ -30,24 +30,24 @@
30
  "120": "NON_PARTICIPATING",
31
  "121": "NON_PARTICIPATING",
32
  "122": "SUCCESS",
33
- "123": "NON_PARTICIPATING",
34
  "124": "NON_PARTICIPATING",
35
- "125": "NON_PARTICIPATING",
36
- "126": "NON_PARTICIPATING",
37
  "127": "NON_PARTICIPATING",
38
  "128": "NON_PARTICIPATING",
39
  "129": "NON_PARTICIPATING",
40
  "13": "SUCCESS",
41
- "130": "NON_PARTICIPATING",
42
  "131": "NON_PARTICIPATING",
43
  "132": "NON_PARTICIPATING",
44
  "133": "NON_PARTICIPATING",
45
- "134": "NON_PARTICIPATING",
46
  "135": "NON_PARTICIPATING",
47
  "136": "NON_PARTICIPATING",
48
  "137": "SUCCESS",
49
- "138": "NON_PARTICIPATING",
50
- "139": "NON_PARTICIPATING",
51
  "14": "NON_PARTICIPATING",
52
  "140": "NON_PARTICIPATING",
53
  "141": "SUCCESS",
@@ -56,14 +56,14 @@
56
  "144": "NON_PARTICIPATING",
57
  "145": "NON_PARTICIPATING",
58
  "146": "NON_PARTICIPATING",
59
- "147": "SUCCESS",
60
  "148": "NON_PARTICIPATING",
61
  "149": "NON_PARTICIPATING",
62
- "15": "SUCCESS",
63
  "150": "NON_PARTICIPATING",
64
  "151": "NON_PARTICIPATING",
65
  "152": "NON_PARTICIPATING",
66
- "153": "SUCCESS",
67
  "154": "NON_PARTICIPATING",
68
  "155": "NON_PARTICIPATING",
69
  "156": "NON_PARTICIPATING",
@@ -75,7 +75,7 @@
75
  "161": "NON_PARTICIPATING",
76
  "162": "NON_PARTICIPATING",
77
  "163": "NON_PARTICIPATING",
78
- "164": "SUCCESS",
79
  "165": "NON_PARTICIPATING",
80
  "166": "NON_PARTICIPATING",
81
  "167": "NON_PARTICIPATING",
@@ -83,13 +83,13 @@
83
  "169": "NON_PARTICIPATING",
84
  "17": "SUCCESS",
85
  "170": "NON_PARTICIPATING",
86
- "171": "NON_PARTICIPATING",
87
  "172": "NON_PARTICIPATING",
88
  "173": "NON_PARTICIPATING",
89
  "174": "NON_PARTICIPATING",
90
- "175": "SUCCESS",
91
  "176": "NON_PARTICIPATING",
92
- "177": "NON_PARTICIPATING",
93
  "178": "NON_PARTICIPATING",
94
  "179": "NON_PARTICIPATING",
95
  "18": "SUCCESS",
@@ -106,19 +106,19 @@
106
  "19": "NON_PARTICIPATING",
107
  "190": "NON_PARTICIPATING",
108
  "191": "NON_PARTICIPATING",
109
- "192": "FAIL",
110
- "193": "SUCCESS",
111
  "194": "NON_PARTICIPATING",
112
  "195": "NON_PARTICIPATING",
113
- "196": "SUCCESS",
114
  "197": "NON_PARTICIPATING",
115
  "198": "NON_PARTICIPATING",
116
  "199": "NON_PARTICIPATING",
117
  "2": "SUCCESS",
118
- "20": "NON_PARTICIPATING",
119
  "200": "NON_PARTICIPATING",
120
  "201": "NON_PARTICIPATING",
121
- "202": "NON_PARTICIPATING",
122
  "203": "NON_PARTICIPATING",
123
  "204": "NON_PARTICIPATING",
124
  "205": "NON_PARTICIPATING",
@@ -127,46 +127,46 @@
127
  "208": "NON_PARTICIPATING",
128
  "209": "NON_PARTICIPATING",
129
  "21": "SUCCESS",
130
- "210": "SUCCESS",
131
  "211": "NON_PARTICIPATING",
132
- "212": "SUCCESS",
133
  "213": "NON_PARTICIPATING",
134
  "214": "NON_PARTICIPATING",
135
  "215": "NON_PARTICIPATING",
136
  "216": "SUCCESS",
137
  "217": "NON_PARTICIPATING",
138
  "218": "NON_PARTICIPATING",
139
- "219": "SUCCESS",
140
  "22": "SUCCESS",
141
  "220": "NON_PARTICIPATING",
142
  "221": "NON_PARTICIPATING",
143
  "222": "NON_PARTICIPATING",
144
  "223": "NON_PARTICIPATING",
145
  "224": "NON_PARTICIPATING",
146
- "225": "SUCCESS",
147
  "226": "NON_PARTICIPATING",
148
  "227": "NON_PARTICIPATING",
149
- "228": "SUCCESS",
150
  "229": "NON_PARTICIPATING",
151
  "23": "SUCCESS",
152
  "230": "SUCCESS",
153
  "231": "NON_PARTICIPATING",
154
  "232": "NON_PARTICIPATING",
155
- "233": "NON_PARTICIPATING",
156
  "234": "NON_PARTICIPATING",
157
  "235": "NON_PARTICIPATING",
158
- "236": "SUCCESS",
159
  "237": "NON_PARTICIPATING",
160
  "238": "NON_PARTICIPATING",
161
- "239": "SUCCESS",
162
  "24": "SUCCESS",
163
  "240": "NON_PARTICIPATING",
164
  "241": "NON_PARTICIPATING",
165
  "242": "NON_PARTICIPATING",
166
- "243": "NON_PARTICIPATING",
167
- "244": "NON_PARTICIPATING",
168
  "245": "NON_PARTICIPATING",
169
- "246": "NON_PARTICIPATING",
170
  "247": "NON_PARTICIPATING",
171
  "248": "NON_PARTICIPATING",
172
  "249": "NON_PARTICIPATING",
@@ -175,7 +175,7 @@
175
  "251": "NON_PARTICIPATING",
176
  "252": "NON_PARTICIPATING",
177
  "253": "NON_PARTICIPATING",
178
- "254": "NON_PARTICIPATING",
179
  "255": "NON_PARTICIPATING",
180
  "26": "SUCCESS",
181
  "27": "SUCCESS",
@@ -188,12 +188,12 @@
188
  "33": "NON_PARTICIPATING",
189
  "34": "NON_PARTICIPATING",
190
  "35": "SUCCESS",
191
- "36": "SUCCESS",
192
  "37": "SUCCESS",
193
  "38": "SUCCESS",
194
  "39": "SUCCESS",
195
- "4": "SUCCESS",
196
- "40": "NON_PARTICIPATING",
197
  "41": "SUCCESS",
198
  "42": "SUCCESS",
199
  "43": "SUCCESS",
@@ -219,10 +219,10 @@
219
  "61": "SUCCESS",
220
  "62": "SUCCESS",
221
  "63": "SUCCESS",
222
- "64": "NON_PARTICIPATING",
223
  "65": "NON_PARTICIPATING",
224
  "66": "NON_PARTICIPATING",
225
- "67": "SUCCESS",
226
  "68": "SUCCESS",
227
  "69": "SUCCESS",
228
  "7": "SUCCESS",
@@ -236,20 +236,20 @@
236
  "77": "SUCCESS",
237
  "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
- "8": "NON_PARTICIPATING",
240
- "80": "SUCCESS",
241
  "81": "SUCCESS",
242
  "82": "SUCCESS",
243
  "83": "SUCCESS",
244
  "84": "SUCCESS",
245
- "85": "NON_PARTICIPATING",
246
  "86": "NON_PARTICIPATING",
247
  "87": "SUCCESS",
248
  "88": "SUCCESS",
249
  "89": "SUCCESS",
250
  "9": "SUCCESS",
251
  "90": "SUCCESS",
252
- "91": "NON_PARTICIPATING",
253
  "92": "SUCCESS",
254
  "93": "NON_PARTICIPATING",
255
  "94": "SUCCESS",
@@ -275,7 +275,7 @@
275
  "initializer_range": 0.02,
276
  "inner_step": 0,
277
  "inner_steps": 0,
278
- "last_allreduce_block": 5639485,
279
  "layer_norm_epsilon": 1e-05,
280
  "model_type": "gpt_optimized",
281
  "n_embd": 1280,
 
1
  {
2
+ "_name_or_path": "AussieAck/DTS25_model",
3
  "activation_function": "gelu_new",
4
  "all_reduce_scores": {
5
  "0": "NON_PARTICIPATING",
6
  "1": "SUCCESS",
7
  "10": "NON_PARTICIPATING",
8
  "100": "NON_PARTICIPATING",
9
+ "101": "SUCCESS",
10
+ "102": "SUCCESS",
11
  "103": "NON_PARTICIPATING",
12
  "104": "SUCCESS",
13
+ "105": "SUCCESS",
14
  "106": "NON_PARTICIPATING",
15
+ "107": "NON_PARTICIPATING",
16
  "108": "NON_PARTICIPATING",
17
  "109": "NON_PARTICIPATING",
18
+ "11": "SUCCESS",
19
+ "110": "SUCCESS",
20
+ "111": "NON_PARTICIPATING",
21
  "112": "NON_PARTICIPATING",
22
  "113": "NON_PARTICIPATING",
23
  "114": "NON_PARTICIPATING",
 
30
  "120": "NON_PARTICIPATING",
31
  "121": "NON_PARTICIPATING",
32
  "122": "SUCCESS",
33
+ "123": "SUCCESS",
34
  "124": "NON_PARTICIPATING",
35
+ "125": "SUCCESS",
36
+ "126": "SUCCESS",
37
  "127": "NON_PARTICIPATING",
38
  "128": "NON_PARTICIPATING",
39
  "129": "NON_PARTICIPATING",
40
  "13": "SUCCESS",
41
+ "130": "SUCCESS",
42
  "131": "NON_PARTICIPATING",
43
  "132": "NON_PARTICIPATING",
44
  "133": "NON_PARTICIPATING",
45
+ "134": "SUCCESS",
46
  "135": "NON_PARTICIPATING",
47
  "136": "NON_PARTICIPATING",
48
  "137": "SUCCESS",
49
+ "138": "SUCCESS",
50
+ "139": "SUCCESS",
51
  "14": "NON_PARTICIPATING",
52
  "140": "NON_PARTICIPATING",
53
  "141": "SUCCESS",
 
56
  "144": "NON_PARTICIPATING",
57
  "145": "NON_PARTICIPATING",
58
  "146": "NON_PARTICIPATING",
59
+ "147": "NON_PARTICIPATING",
60
  "148": "NON_PARTICIPATING",
61
  "149": "NON_PARTICIPATING",
62
+ "15": "NON_PARTICIPATING",
63
  "150": "NON_PARTICIPATING",
64
  "151": "NON_PARTICIPATING",
65
  "152": "NON_PARTICIPATING",
66
+ "153": "NON_PARTICIPATING",
67
  "154": "NON_PARTICIPATING",
68
  "155": "NON_PARTICIPATING",
69
  "156": "NON_PARTICIPATING",
 
75
  "161": "NON_PARTICIPATING",
76
  "162": "NON_PARTICIPATING",
77
  "163": "NON_PARTICIPATING",
78
+ "164": "NON_PARTICIPATING",
79
  "165": "NON_PARTICIPATING",
80
  "166": "NON_PARTICIPATING",
81
  "167": "NON_PARTICIPATING",
 
83
  "169": "NON_PARTICIPATING",
84
  "17": "SUCCESS",
85
  "170": "NON_PARTICIPATING",
86
+ "171": "SUCCESS",
87
  "172": "NON_PARTICIPATING",
88
  "173": "NON_PARTICIPATING",
89
  "174": "NON_PARTICIPATING",
90
+ "175": "NON_PARTICIPATING",
91
  "176": "NON_PARTICIPATING",
92
+ "177": "SUCCESS",
93
  "178": "NON_PARTICIPATING",
94
  "179": "NON_PARTICIPATING",
95
  "18": "SUCCESS",
 
106
  "19": "NON_PARTICIPATING",
107
  "190": "NON_PARTICIPATING",
108
  "191": "NON_PARTICIPATING",
109
+ "192": "NON_PARTICIPATING",
110
+ "193": "NON_PARTICIPATING",
111
  "194": "NON_PARTICIPATING",
112
  "195": "NON_PARTICIPATING",
113
+ "196": "NON_PARTICIPATING",
114
  "197": "NON_PARTICIPATING",
115
  "198": "NON_PARTICIPATING",
116
  "199": "NON_PARTICIPATING",
117
  "2": "SUCCESS",
118
+ "20": "SUCCESS",
119
  "200": "NON_PARTICIPATING",
120
  "201": "NON_PARTICIPATING",
121
+ "202": "SUCCESS",
122
  "203": "NON_PARTICIPATING",
123
  "204": "NON_PARTICIPATING",
124
  "205": "NON_PARTICIPATING",
 
127
  "208": "NON_PARTICIPATING",
128
  "209": "NON_PARTICIPATING",
129
  "21": "SUCCESS",
130
+ "210": "NON_PARTICIPATING",
131
  "211": "NON_PARTICIPATING",
132
+ "212": "NON_PARTICIPATING",
133
  "213": "NON_PARTICIPATING",
134
  "214": "NON_PARTICIPATING",
135
  "215": "NON_PARTICIPATING",
136
  "216": "SUCCESS",
137
  "217": "NON_PARTICIPATING",
138
  "218": "NON_PARTICIPATING",
139
+ "219": "NON_PARTICIPATING",
140
  "22": "SUCCESS",
141
  "220": "NON_PARTICIPATING",
142
  "221": "NON_PARTICIPATING",
143
  "222": "NON_PARTICIPATING",
144
  "223": "NON_PARTICIPATING",
145
  "224": "NON_PARTICIPATING",
146
+ "225": "NON_PARTICIPATING",
147
  "226": "NON_PARTICIPATING",
148
  "227": "NON_PARTICIPATING",
149
+ "228": "NON_PARTICIPATING",
150
  "229": "NON_PARTICIPATING",
151
  "23": "SUCCESS",
152
  "230": "SUCCESS",
153
  "231": "NON_PARTICIPATING",
154
  "232": "NON_PARTICIPATING",
155
+ "233": "SUCCESS",
156
  "234": "NON_PARTICIPATING",
157
  "235": "NON_PARTICIPATING",
158
+ "236": "NON_PARTICIPATING",
159
  "237": "NON_PARTICIPATING",
160
  "238": "NON_PARTICIPATING",
161
+ "239": "NON_PARTICIPATING",
162
  "24": "SUCCESS",
163
  "240": "NON_PARTICIPATING",
164
  "241": "NON_PARTICIPATING",
165
  "242": "NON_PARTICIPATING",
166
+ "243": "SUCCESS",
167
+ "244": "SUCCESS",
168
  "245": "NON_PARTICIPATING",
169
+ "246": "SUCCESS",
170
  "247": "NON_PARTICIPATING",
171
  "248": "NON_PARTICIPATING",
172
  "249": "NON_PARTICIPATING",
 
175
  "251": "NON_PARTICIPATING",
176
  "252": "NON_PARTICIPATING",
177
  "253": "NON_PARTICIPATING",
178
+ "254": "SUCCESS",
179
  "255": "NON_PARTICIPATING",
180
  "26": "SUCCESS",
181
  "27": "SUCCESS",
 
188
  "33": "NON_PARTICIPATING",
189
  "34": "NON_PARTICIPATING",
190
  "35": "SUCCESS",
191
+ "36": "NON_PARTICIPATING",
192
  "37": "SUCCESS",
193
  "38": "SUCCESS",
194
  "39": "SUCCESS",
195
+ "4": "NON_PARTICIPATING",
196
+ "40": "SUCCESS",
197
  "41": "SUCCESS",
198
  "42": "SUCCESS",
199
  "43": "SUCCESS",
 
219
  "61": "SUCCESS",
220
  "62": "SUCCESS",
221
  "63": "SUCCESS",
222
+ "64": "SUCCESS",
223
  "65": "NON_PARTICIPATING",
224
  "66": "NON_PARTICIPATING",
225
+ "67": "NON_PARTICIPATING",
226
  "68": "SUCCESS",
227
  "69": "SUCCESS",
228
  "7": "SUCCESS",
 
236
  "77": "SUCCESS",
237
  "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
+ "8": "SUCCESS",
240
+ "80": "NON_PARTICIPATING",
241
  "81": "SUCCESS",
242
  "82": "SUCCESS",
243
  "83": "SUCCESS",
244
  "84": "SUCCESS",
245
+ "85": "SUCCESS",
246
  "86": "NON_PARTICIPATING",
247
  "87": "SUCCESS",
248
  "88": "SUCCESS",
249
  "89": "SUCCESS",
250
  "9": "SUCCESS",
251
  "90": "SUCCESS",
252
+ "91": "FAIL",
253
  "92": "SUCCESS",
254
  "93": "NON_PARTICIPATING",
255
  "94": "SUCCESS",
 
275
  "initializer_range": 0.02,
276
  "inner_step": 0,
277
  "inner_steps": 0,
278
+ "last_allreduce_block": 5641040,
279
  "layer_norm_epsilon": 1e-05,
280
  "model_type": "gpt_optimized",
281
  "n_embd": 1280,
inner_optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9344c0edc423a109b93625e53e045c7dc45ffb1350613ecdb978d03da3da0f20
3
  size 8081782026
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c484183a39085e0a0231cef1ef9dac69e536f4b8e30976bc8f578d85e9f63e9
3
  size 8081782026
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:594c2c675881510bb478dcd1fee5f73aabdb7e37f7a060c16be9e2d71175f901
3
  size 4040701744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b72f2f4cfa96d22e9e515795f3830a5a9132f18917c43dd54918719f1778ee8
3
  size 4040701744
outer_optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b84fc41a85ab886f853c5e46189c392af2e1e7076523d19a5ffb8accdf07153e
3
  size 4040805354
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:220893adc425d15fa7f6bdfd3bd7a52ffb8a6ea4c365d3500d0073703bdad851
3
  size 4040805354