wandb02 commited on
Commit
60e7449
·
verified ·
1 Parent(s): d4a2a27

Run 4. Outer Step 13. Inner Step 0.

Browse files
Files changed (3) hide show
  1. config.json +75 -75
  2. inner_optimizer.pt +1 -1
  3. model.safetensors +1 -1
config.json CHANGED
@@ -3,7 +3,7 @@
3
  "activation_function": "gelu_new",
4
  "all_reduce_scores": {
5
  "0": "NON_PARTICIPATING",
6
- "1": "NON_PARTICIPATING",
7
  "10": "SUCCESS",
8
  "100": "NON_PARTICIPATING",
9
  "101": "NON_PARTICIPATING",
@@ -11,38 +11,38 @@
11
  "103": "NON_PARTICIPATING",
12
  "104": "NON_PARTICIPATING",
13
  "105": "NON_PARTICIPATING",
14
- "106": "NON_PARTICIPATING",
15
  "107": "SUCCESS",
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",
24
  "115": "NON_PARTICIPATING",
25
  "116": "NON_PARTICIPATING",
26
- "117": "SUCCESS",
27
- "118": "SUCCESS",
28
  "119": "NON_PARTICIPATING",
29
- "12": "NON_PARTICIPATING",
30
  "120": "SUCCESS",
31
  "121": "NON_PARTICIPATING",
32
- "122": "SUCCESS",
33
- "123": "SUCCESS",
34
  "124": "NON_PARTICIPATING",
35
- "125": "SUCCESS",
36
  "126": "NON_PARTICIPATING",
37
- "127": "NON_PARTICIPATING",
38
  "128": "NON_PARTICIPATING",
39
- "129": "NON_PARTICIPATING",
40
  "13": "NON_PARTICIPATING",
41
  "130": "NON_PARTICIPATING",
42
  "131": "NON_PARTICIPATING",
43
  "132": "SUCCESS",
44
  "133": "NON_PARTICIPATING",
45
- "134": "SUCCESS",
46
  "135": "SUCCESS",
47
  "136": "NON_PARTICIPATING",
48
  "137": "NON_PARTICIPATING",
@@ -51,25 +51,25 @@
51
  "14": "NON_PARTICIPATING",
52
  "140": "NON_PARTICIPATING",
53
  "141": "NON_PARTICIPATING",
54
- "142": "NON_PARTICIPATING",
55
  "143": "NON_PARTICIPATING",
56
  "144": "NON_PARTICIPATING",
57
  "145": "NON_PARTICIPATING",
58
  "146": "NON_PARTICIPATING",
59
- "147": "NON_PARTICIPATING",
60
  "148": "SUCCESS",
61
  "149": "SUCCESS",
62
  "15": "SUCCESS",
63
- "150": "SUCCESS",
64
  "151": "NON_PARTICIPATING",
65
  "152": "NON_PARTICIPATING",
66
- "153": "SUCCESS",
67
  "154": "SUCCESS",
68
- "155": "NON_PARTICIPATING",
69
  "156": "SUCCESS",
70
  "157": "NON_PARTICIPATING",
71
  "158": "NON_PARTICIPATING",
72
- "159": "SUCCESS",
73
  "16": "NON_PARTICIPATING",
74
  "160": "SUCCESS",
75
  "161": "NON_PARTICIPATING",
@@ -78,19 +78,19 @@
78
  "164": "NON_PARTICIPATING",
79
  "165": "NON_PARTICIPATING",
80
  "166": "NON_PARTICIPATING",
81
- "167": "SUCCESS",
82
  "168": "NON_PARTICIPATING",
83
  "169": "NON_PARTICIPATING",
84
- "17": "NON_PARTICIPATING",
85
- "170": "NON_PARTICIPATING",
86
  "171": "NON_PARTICIPATING",
87
  "172": "SUCCESS",
88
  "173": "NON_PARTICIPATING",
89
  "174": "NON_PARTICIPATING",
90
- "175": "SUCCESS",
91
  "176": "NON_PARTICIPATING",
92
- "177": "SUCCESS",
93
- "178": "SUCCESS",
94
  "179": "NON_PARTICIPATING",
95
  "18": "NON_PARTICIPATING",
96
  "180": "NON_PARTICIPATING",
@@ -100,29 +100,29 @@
100
  "184": "NON_PARTICIPATING",
101
  "185": "NON_PARTICIPATING",
102
  "186": "NON_PARTICIPATING",
103
- "187": "NON_PARTICIPATING",
104
  "188": "NON_PARTICIPATING",
105
  "189": "NON_PARTICIPATING",
106
  "19": "NON_PARTICIPATING",
107
  "190": "NON_PARTICIPATING",
108
  "191": "NON_PARTICIPATING",
109
- "192": "SUCCESS",
110
  "193": "NON_PARTICIPATING",
111
- "194": "SUCCESS",
112
  "195": "SUCCESS",
113
  "196": "NON_PARTICIPATING",
114
  "197": "NON_PARTICIPATING",
115
  "198": "NON_PARTICIPATING",
116
  "199": "NON_PARTICIPATING",
117
- "2": "NON_PARTICIPATING",
118
- "20": "NON_PARTICIPATING",
119
- "200": "NON_PARTICIPATING",
120
- "201": "NON_PARTICIPATING",
121
- "202": "SUCCESS",
122
- "203": "SUCCESS",
123
  "204": "NON_PARTICIPATING",
124
  "205": "NON_PARTICIPATING",
125
- "206": "SUCCESS",
126
  "207": "NON_PARTICIPATING",
127
  "208": "NON_PARTICIPATING",
128
  "209": "NON_PARTICIPATING",
@@ -132,25 +132,25 @@
132
  "212": "NON_PARTICIPATING",
133
  "213": "NON_PARTICIPATING",
134
  "214": "SUCCESS",
135
- "215": "SUCCESS",
136
- "216": "SUCCESS",
137
  "217": "NON_PARTICIPATING",
138
  "218": "SUCCESS",
139
  "219": "NON_PARTICIPATING",
140
- "22": "SUCCESS",
141
  "220": "NON_PARTICIPATING",
142
  "221": "NON_PARTICIPATING",
143
- "222": "SUCCESS",
144
  "223": "NON_PARTICIPATING",
145
  "224": "NON_PARTICIPATING",
146
- "225": "NON_PARTICIPATING",
147
  "226": "SUCCESS",
148
  "227": "NON_PARTICIPATING",
149
  "228": "NON_PARTICIPATING",
150
  "229": "NON_PARTICIPATING",
151
  "23": "SUCCESS",
152
  "230": "NON_PARTICIPATING",
153
- "231": "NON_PARTICIPATING",
154
  "232": "NON_PARTICIPATING",
155
  "233": "NON_PARTICIPATING",
156
  "234": "NON_PARTICIPATING",
@@ -167,62 +167,62 @@
167
  "244": "NON_PARTICIPATING",
168
  "245": "NON_PARTICIPATING",
169
  "246": "NON_PARTICIPATING",
170
- "247": "SUCCESS",
171
  "248": "NON_PARTICIPATING",
172
- "249": "SUCCESS",
173
  "25": "SUCCESS",
174
- "250": "NON_PARTICIPATING",
175
  "251": "NON_PARTICIPATING",
176
  "252": "NON_PARTICIPATING",
177
  "253": "NON_PARTICIPATING",
178
  "254": "NON_PARTICIPATING",
179
- "255": "SUCCESS",
180
  "26": "SUCCESS",
181
  "27": "SUCCESS",
182
  "28": "NON_PARTICIPATING",
183
  "29": "SUCCESS",
184
  "3": "NON_PARTICIPATING",
185
- "30": "NON_PARTICIPATING",
186
  "31": "NON_PARTICIPATING",
187
  "32": "SUCCESS",
188
  "33": "NON_PARTICIPATING",
189
  "34": "SUCCESS",
190
- "35": "SUCCESS",
191
  "36": "NON_PARTICIPATING",
192
- "37": "NON_PARTICIPATING",
193
  "38": "NON_PARTICIPATING",
194
  "39": "NON_PARTICIPATING",
195
- "4": "SUCCESS",
196
- "40": "SUCCESS",
197
  "41": "NON_PARTICIPATING",
198
  "42": "NON_PARTICIPATING",
199
  "43": "NON_PARTICIPATING",
200
  "44": "NON_PARTICIPATING",
201
  "45": "NON_PARTICIPATING",
202
- "46": "NON_PARTICIPATING",
203
- "47": "NON_PARTICIPATING",
204
  "48": "NON_PARTICIPATING",
205
  "49": "SUCCESS",
206
  "5": "NON_PARTICIPATING",
207
- "50": "NON_PARTICIPATING",
208
  "51": "SUCCESS",
209
  "52": "NON_PARTICIPATING",
210
  "53": "NON_PARTICIPATING",
211
  "54": "NON_PARTICIPATING",
212
  "55": "SUCCESS",
213
- "56": "NON_PARTICIPATING",
214
- "57": "NON_PARTICIPATING",
215
- "58": "SUCCESS",
216
  "59": "NON_PARTICIPATING",
217
  "6": "SUCCESS",
218
  "60": "NON_PARTICIPATING",
219
  "61": "NON_PARTICIPATING",
220
  "62": "NON_PARTICIPATING",
221
  "63": "SUCCESS",
222
- "64": "NON_PARTICIPATING",
223
  "65": "NON_PARTICIPATING",
224
  "66": "NON_PARTICIPATING",
225
- "67": "SUCCESS",
226
  "68": "NON_PARTICIPATING",
227
  "69": "NON_PARTICIPATING",
228
  "7": "NON_PARTICIPATING",
@@ -234,27 +234,27 @@
234
  "75": "SUCCESS",
235
  "76": "NON_PARTICIPATING",
236
  "77": "NON_PARTICIPATING",
237
- "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
- "8": "NON_PARTICIPATING",
240
  "80": "SUCCESS",
241
  "81": "NON_PARTICIPATING",
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
244
  "84": "NON_PARTICIPATING",
245
- "85": "NON_PARTICIPATING",
246
  "86": "NON_PARTICIPATING",
247
  "87": "SUCCESS",
248
  "88": "SUCCESS",
249
- "89": "SUCCESS",
250
  "9": "NON_PARTICIPATING",
251
- "90": "SUCCESS",
252
- "91": "NON_PARTICIPATING",
253
- "92": "SUCCESS",
254
  "93": "NON_PARTICIPATING",
255
  "94": "NON_PARTICIPATING",
256
- "95": "SUCCESS",
257
- "96": "NON_PARTICIPATING",
258
  "97": "NON_PARTICIPATING",
259
  "98": "NON_PARTICIPATING",
260
  "99": "SUCCESS"
@@ -268,20 +268,20 @@
268
  "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
269
  },
270
  "block_list": [
271
- 5506763,
272
- 5506767,
273
- 5506771,
274
- 5506775,
275
- 5506779
276
  ],
277
  "block_size": 1024,
278
  "bos_token_id": 50256,
279
  "embd_pdrop": 0.1,
280
  "eos_token_id": 50256,
281
  "initializer_range": 0.02,
282
- "inner_step": 103,
283
  "inner_steps": 0,
284
- "last_allreduce_block": 5503964,
285
  "layer_norm_epsilon": 1e-05,
286
  "model_type": "gpt_optimized",
287
  "n_embd": 1280,
 
3
  "activation_function": "gelu_new",
4
  "all_reduce_scores": {
5
  "0": "NON_PARTICIPATING",
6
+ "1": "SUCCESS",
7
  "10": "SUCCESS",
8
  "100": "NON_PARTICIPATING",
9
  "101": "NON_PARTICIPATING",
 
11
  "103": "NON_PARTICIPATING",
12
  "104": "NON_PARTICIPATING",
13
  "105": "NON_PARTICIPATING",
14
+ "106": "SUCCESS",
15
  "107": "SUCCESS",
16
  "108": "NON_PARTICIPATING",
17
+ "109": "SUCCESS",
18
+ "11": "NON_PARTICIPATING",
19
  "110": "SUCCESS",
20
  "111": "NON_PARTICIPATING",
21
  "112": "NON_PARTICIPATING",
22
  "113": "NON_PARTICIPATING",
23
+ "114": "SUCCESS",
24
  "115": "NON_PARTICIPATING",
25
  "116": "NON_PARTICIPATING",
26
+ "117": "NON_PARTICIPATING",
27
+ "118": "NON_PARTICIPATING",
28
  "119": "NON_PARTICIPATING",
29
+ "12": "SUCCESS",
30
  "120": "SUCCESS",
31
  "121": "NON_PARTICIPATING",
32
+ "122": "NON_PARTICIPATING",
33
+ "123": "NON_PARTICIPATING",
34
  "124": "NON_PARTICIPATING",
35
+ "125": "NON_PARTICIPATING",
36
  "126": "NON_PARTICIPATING",
37
+ "127": "SUCCESS",
38
  "128": "NON_PARTICIPATING",
39
+ "129": "SUCCESS",
40
  "13": "NON_PARTICIPATING",
41
  "130": "NON_PARTICIPATING",
42
  "131": "NON_PARTICIPATING",
43
  "132": "SUCCESS",
44
  "133": "NON_PARTICIPATING",
45
+ "134": "NON_PARTICIPATING",
46
  "135": "SUCCESS",
47
  "136": "NON_PARTICIPATING",
48
  "137": "NON_PARTICIPATING",
 
51
  "14": "NON_PARTICIPATING",
52
  "140": "NON_PARTICIPATING",
53
  "141": "NON_PARTICIPATING",
54
+ "142": "SUCCESS",
55
  "143": "NON_PARTICIPATING",
56
  "144": "NON_PARTICIPATING",
57
  "145": "NON_PARTICIPATING",
58
  "146": "NON_PARTICIPATING",
59
+ "147": "SUCCESS",
60
  "148": "SUCCESS",
61
  "149": "SUCCESS",
62
  "15": "SUCCESS",
63
+ "150": "NON_PARTICIPATING",
64
  "151": "NON_PARTICIPATING",
65
  "152": "NON_PARTICIPATING",
66
+ "153": "NON_PARTICIPATING",
67
  "154": "SUCCESS",
68
+ "155": "SUCCESS",
69
  "156": "SUCCESS",
70
  "157": "NON_PARTICIPATING",
71
  "158": "NON_PARTICIPATING",
72
+ "159": "NON_PARTICIPATING",
73
  "16": "NON_PARTICIPATING",
74
  "160": "SUCCESS",
75
  "161": "NON_PARTICIPATING",
 
78
  "164": "NON_PARTICIPATING",
79
  "165": "NON_PARTICIPATING",
80
  "166": "NON_PARTICIPATING",
81
+ "167": "NON_PARTICIPATING",
82
  "168": "NON_PARTICIPATING",
83
  "169": "NON_PARTICIPATING",
84
+ "17": "SUCCESS",
85
+ "170": "SUCCESS",
86
  "171": "NON_PARTICIPATING",
87
  "172": "SUCCESS",
88
  "173": "NON_PARTICIPATING",
89
  "174": "NON_PARTICIPATING",
90
+ "175": "NON_PARTICIPATING",
91
  "176": "NON_PARTICIPATING",
92
+ "177": "NON_PARTICIPATING",
93
+ "178": "NON_PARTICIPATING",
94
  "179": "NON_PARTICIPATING",
95
  "18": "NON_PARTICIPATING",
96
  "180": "NON_PARTICIPATING",
 
100
  "184": "NON_PARTICIPATING",
101
  "185": "NON_PARTICIPATING",
102
  "186": "NON_PARTICIPATING",
103
+ "187": "SUCCESS",
104
  "188": "NON_PARTICIPATING",
105
  "189": "NON_PARTICIPATING",
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": "SUCCESS",
113
  "196": "NON_PARTICIPATING",
114
  "197": "NON_PARTICIPATING",
115
  "198": "NON_PARTICIPATING",
116
  "199": "NON_PARTICIPATING",
117
+ "2": "SUCCESS",
118
+ "20": "SUCCESS",
119
+ "200": "SUCCESS",
120
+ "201": "SUCCESS",
121
+ "202": "NON_PARTICIPATING",
122
+ "203": "NON_PARTICIPATING",
123
  "204": "NON_PARTICIPATING",
124
  "205": "NON_PARTICIPATING",
125
+ "206": "NON_PARTICIPATING",
126
  "207": "NON_PARTICIPATING",
127
  "208": "NON_PARTICIPATING",
128
  "209": "NON_PARTICIPATING",
 
132
  "212": "NON_PARTICIPATING",
133
  "213": "NON_PARTICIPATING",
134
  "214": "SUCCESS",
135
+ "215": "NON_PARTICIPATING",
136
+ "216": "NON_PARTICIPATING",
137
  "217": "NON_PARTICIPATING",
138
  "218": "SUCCESS",
139
  "219": "NON_PARTICIPATING",
140
+ "22": "NON_PARTICIPATING",
141
  "220": "NON_PARTICIPATING",
142
  "221": "NON_PARTICIPATING",
143
+ "222": "NON_PARTICIPATING",
144
  "223": "NON_PARTICIPATING",
145
  "224": "NON_PARTICIPATING",
146
+ "225": "FAIL",
147
  "226": "SUCCESS",
148
  "227": "NON_PARTICIPATING",
149
  "228": "NON_PARTICIPATING",
150
  "229": "NON_PARTICIPATING",
151
  "23": "SUCCESS",
152
  "230": "NON_PARTICIPATING",
153
+ "231": "SUCCESS",
154
  "232": "NON_PARTICIPATING",
155
  "233": "NON_PARTICIPATING",
156
  "234": "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",
173
  "25": "SUCCESS",
174
+ "250": "SUCCESS",
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",
182
  "28": "NON_PARTICIPATING",
183
  "29": "SUCCESS",
184
  "3": "NON_PARTICIPATING",
185
+ "30": "SUCCESS",
186
  "31": "NON_PARTICIPATING",
187
  "32": "SUCCESS",
188
  "33": "NON_PARTICIPATING",
189
  "34": "SUCCESS",
190
+ "35": "NON_PARTICIPATING",
191
  "36": "NON_PARTICIPATING",
192
+ "37": "SUCCESS",
193
  "38": "NON_PARTICIPATING",
194
  "39": "NON_PARTICIPATING",
195
+ "4": "NON_PARTICIPATING",
196
+ "40": "NON_PARTICIPATING",
197
  "41": "NON_PARTICIPATING",
198
  "42": "NON_PARTICIPATING",
199
  "43": "NON_PARTICIPATING",
200
  "44": "NON_PARTICIPATING",
201
  "45": "NON_PARTICIPATING",
202
+ "46": "SUCCESS",
203
+ "47": "SUCCESS",
204
  "48": "NON_PARTICIPATING",
205
  "49": "SUCCESS",
206
  "5": "NON_PARTICIPATING",
207
+ "50": "SUCCESS",
208
  "51": "SUCCESS",
209
  "52": "NON_PARTICIPATING",
210
  "53": "NON_PARTICIPATING",
211
  "54": "NON_PARTICIPATING",
212
  "55": "SUCCESS",
213
+ "56": "SUCCESS",
214
+ "57": "SUCCESS",
215
+ "58": "NON_PARTICIPATING",
216
  "59": "NON_PARTICIPATING",
217
  "6": "SUCCESS",
218
  "60": "NON_PARTICIPATING",
219
  "61": "NON_PARTICIPATING",
220
  "62": "NON_PARTICIPATING",
221
  "63": "SUCCESS",
222
+ "64": "FAIL",
223
  "65": "NON_PARTICIPATING",
224
  "66": "NON_PARTICIPATING",
225
+ "67": "NON_PARTICIPATING",
226
  "68": "NON_PARTICIPATING",
227
  "69": "NON_PARTICIPATING",
228
  "7": "NON_PARTICIPATING",
 
234
  "75": "SUCCESS",
235
  "76": "NON_PARTICIPATING",
236
  "77": "NON_PARTICIPATING",
237
+ "78": "NON_PARTICIPATING",
238
  "79": "NON_PARTICIPATING",
239
+ "8": "SUCCESS",
240
  "80": "SUCCESS",
241
  "81": "NON_PARTICIPATING",
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
244
  "84": "NON_PARTICIPATING",
245
+ "85": "SUCCESS",
246
  "86": "NON_PARTICIPATING",
247
  "87": "SUCCESS",
248
  "88": "SUCCESS",
249
+ "89": "NON_PARTICIPATING",
250
  "9": "NON_PARTICIPATING",
251
+ "90": "NON_PARTICIPATING",
252
+ "91": "SUCCESS",
253
+ "92": "NON_PARTICIPATING",
254
  "93": "NON_PARTICIPATING",
255
  "94": "NON_PARTICIPATING",
256
+ "95": "NON_PARTICIPATING",
257
+ "96": "SUCCESS",
258
  "97": "NON_PARTICIPATING",
259
  "98": "NON_PARTICIPATING",
260
  "99": "SUCCESS"
 
268
  "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
269
  },
270
  "block_list": [
271
+ 5510036,
272
+ 5510040,
273
+ 5510044,
274
+ 5510047,
275
+ 5510051
276
  ],
277
  "block_size": 1024,
278
  "bos_token_id": 50256,
279
  "embd_pdrop": 0.1,
280
  "eos_token_id": 50256,
281
  "initializer_range": 0.02,
282
+ "inner_step": 0,
283
  "inner_steps": 0,
284
+ "last_allreduce_block": 5509976,
285
  "layer_norm_epsilon": 1e-05,
286
  "model_type": "gpt_optimized",
287
  "n_embd": 1280,
inner_optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b4174df9157806e995af421c01ce0f2435b9304c44070ade81238a46c0c89e4d
3
  size 8081782503
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8ec4794f601cb174cf272d6f2199a1a5d3ea2fe330f57a0334782a9ce74fdda2
3
  size 8081782503
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:db696b6d8f04ccbd048432cfc478d5a9cc21d5ef56c7f5fce6a5ecc5fff3d084
3
  size 4040701744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4758b9afba5f425a99f991687d041c96d8c8b6b6fa323847fe8b1a1b849d3f84
3
  size 4040701744