zeras141a commited on
Commit
6316d0d
·
verified ·
1 Parent(s): e41a65b

Run 4. Outer Step 48. Inner Step 0.

Browse files
Files changed (3) hide show
  1. config.json +46 -65
  2. inner_optimizer.pt +1 -1
  3. model.safetensors +1 -1
config.json CHANGED
@@ -11,11 +11,11 @@
11
  "103": "NON_PARTICIPATING",
12
  "104": "NON_PARTICIPATING",
13
  "105": "NON_PARTICIPATING",
14
- "106": "SUCCESS",
15
  "107": "NON_PARTICIPATING",
16
  "108": "NON_PARTICIPATING",
17
  "109": "NON_PARTICIPATING",
18
- "11": "NON_PARTICIPATING",
19
  "110": "NON_PARTICIPATING",
20
  "111": "NON_PARTICIPATING",
21
  "112": "NON_PARTICIPATING",
@@ -30,15 +30,15 @@
30
  "120": "NON_PARTICIPATING",
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": "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",
@@ -48,7 +48,7 @@
48
  "137": "NON_PARTICIPATING",
49
  "138": "NON_PARTICIPATING",
50
  "139": "NON_PARTICIPATING",
51
- "14": "SUCCESS",
52
  "140": "NON_PARTICIPATING",
53
  "141": "NON_PARTICIPATING",
54
  "142": "NON_PARTICIPATING",
@@ -57,33 +57,33 @@
57
  "145": "NON_PARTICIPATING",
58
  "146": "NON_PARTICIPATING",
59
  "147": "NON_PARTICIPATING",
60
- "148": "SUCCESS",
61
- "149": "NON_PARTICIPATING",
62
- "15": "FAIL",
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",
70
  "157": "SUCCESS",
71
  "158": "NON_PARTICIPATING",
72
- "159": "SUCCESS",
73
- "16": "FAIL",
74
  "160": "NON_PARTICIPATING",
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",
82
- "168": "SUCCESS",
83
- "169": "SUCCESS",
84
  "17": "SUCCESS",
85
- "170": "NON_PARTICIPATING",
86
- "171": "FAIL",
87
  "172": "NON_PARTICIPATING",
88
  "173": "NON_PARTICIPATING",
89
  "174": "NON_PARTICIPATING",
@@ -93,19 +93,19 @@
93
  "178": "NON_PARTICIPATING",
94
  "179": "NON_PARTICIPATING",
95
  "18": "SUCCESS",
96
- "180": "NON_PARTICIPATING",
97
  "181": "NON_PARTICIPATING",
98
  "182": "NON_PARTICIPATING",
99
  "183": "NON_PARTICIPATING",
100
- "184": "NON_PARTICIPATING",
101
- "185": "FAIL",
102
  "186": "NON_PARTICIPATING",
103
  "187": "NON_PARTICIPATING",
104
  "188": "NON_PARTICIPATING",
105
  "189": "NON_PARTICIPATING",
106
  "19": "SUCCESS",
107
- "190": "SUCCESS",
108
- "191": "NON_PARTICIPATING",
109
  "192": "NON_PARTICIPATING",
110
  "193": "NON_PARTICIPATING",
111
  "194": "NON_PARTICIPATING",
@@ -118,14 +118,14 @@
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",
125
  "206": "NON_PARTICIPATING",
126
  "207": "NON_PARTICIPATING",
127
  "208": "NON_PARTICIPATING",
128
- "209": "SUCCESS",
129
  "21": "SUCCESS",
130
  "210": "NON_PARTICIPATING",
131
  "211": "NON_PARTICIPATING",
@@ -152,28 +152,28 @@
152
  "230": "NON_PARTICIPATING",
153
  "231": "NON_PARTICIPATING",
154
  "232": "NON_PARTICIPATING",
155
- "233": "FAIL",
156
  "234": "NON_PARTICIPATING",
157
  "235": "NON_PARTICIPATING",
158
- "236": "NON_PARTICIPATING",
159
- "237": "NON_PARTICIPATING",
160
- "238": "SUCCESS",
161
  "239": "NON_PARTICIPATING",
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": "FAIL",
170
  "247": "NON_PARTICIPATING",
171
  "248": "NON_PARTICIPATING",
172
  "249": "NON_PARTICIPATING",
173
  "25": "SUCCESS",
174
- "250": "NON_PARTICIPATING",
175
  "251": "NON_PARTICIPATING",
176
- "252": "FAIL",
177
  "253": "NON_PARTICIPATING",
178
  "254": "NON_PARTICIPATING",
179
  "255": "SUCCESS",
@@ -193,9 +193,9 @@
193
  "38": "SUCCESS",
194
  "39": "SUCCESS",
195
  "4": "SUCCESS",
196
- "40": "FAIL",
197
  "41": "SUCCESS",
198
- "42": "SUCCESS",
199
  "43": "SUCCESS",
200
  "44": "SUCCESS",
201
  "45": "SUCCESS",
@@ -213,7 +213,7 @@
213
  "56": "SUCCESS",
214
  "57": "SUCCESS",
215
  "58": "NON_PARTICIPATING",
216
- "59": "FAIL",
217
  "6": "NON_PARTICIPATING",
218
  "60": "SUCCESS",
219
  "61": "NON_PARTICIPATING",
@@ -237,7 +237,7 @@
237
  "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
  "8": "SUCCESS",
240
- "80": "FAIL",
241
  "81": "SUCCESS",
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
@@ -247,7 +247,7 @@
247
  "87": "SUCCESS",
248
  "88": "SUCCESS",
249
  "89": "SUCCESS",
250
- "9": "NON_PARTICIPATING",
251
  "90": "SUCCESS",
252
  "91": "SUCCESS",
253
  "92": "SUCCESS",
@@ -268,39 +268,20 @@
268
  "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
269
  },
270
  "block_list": [
271
- 5568820,
272
- 5568822,
273
- 5568823,
274
- 5568825,
275
- 5568827,
276
- 5568829,
277
- 5568830,
278
- 5568831,
279
- 5568832,
280
- 5568834,
281
- 5568835,
282
- 5568836,
283
- 5568837,
284
- 5568838,
285
- 5568840,
286
- 5568841,
287
- 5568843,
288
- 5568844,
289
- 5568845,
290
- 5568846,
291
- 5568847,
292
- 5568848,
293
- 5568848,
294
- 5568849
295
  ],
296
  "block_size": 1024,
297
  "bos_token_id": 50256,
298
  "embd_pdrop": 0.1,
299
  "eos_token_id": 50256,
300
  "initializer_range": 0.02,
301
- "inner_step": 67,
302
  "inner_steps": 0,
303
- "last_allreduce_block": 5566279,
304
  "layer_norm_epsilon": 1e-05,
305
  "model_type": "gpt_optimized",
306
  "n_embd": 1280,
 
11
  "103": "NON_PARTICIPATING",
12
  "104": "NON_PARTICIPATING",
13
  "105": "NON_PARTICIPATING",
14
+ "106": "NON_PARTICIPATING",
15
  "107": "NON_PARTICIPATING",
16
  "108": "NON_PARTICIPATING",
17
  "109": "NON_PARTICIPATING",
18
+ "11": "SUCCESS",
19
  "110": "NON_PARTICIPATING",
20
  "111": "NON_PARTICIPATING",
21
  "112": "NON_PARTICIPATING",
 
30
  "120": "NON_PARTICIPATING",
31
  "121": "NON_PARTICIPATING",
32
  "122": "NON_PARTICIPATING",
33
+ "123": "SUCCESS",
34
  "124": "NON_PARTICIPATING",
35
  "125": "NON_PARTICIPATING",
36
  "126": "NON_PARTICIPATING",
37
  "127": "NON_PARTICIPATING",
38
  "128": "NON_PARTICIPATING",
39
+ "129": "SUCCESS",
40
  "13": "SUCCESS",
41
+ "130": "NON_PARTICIPATING",
42
  "131": "NON_PARTICIPATING",
43
  "132": "NON_PARTICIPATING",
44
  "133": "NON_PARTICIPATING",
 
48
  "137": "NON_PARTICIPATING",
49
  "138": "NON_PARTICIPATING",
50
  "139": "NON_PARTICIPATING",
51
+ "14": "NON_PARTICIPATING",
52
  "140": "NON_PARTICIPATING",
53
  "141": "NON_PARTICIPATING",
54
  "142": "NON_PARTICIPATING",
 
57
  "145": "NON_PARTICIPATING",
58
  "146": "NON_PARTICIPATING",
59
  "147": "NON_PARTICIPATING",
60
+ "148": "NON_PARTICIPATING",
61
+ "149": "SUCCESS",
62
+ "15": "SUCCESS",
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": "SUCCESS",
70
  "157": "SUCCESS",
71
  "158": "NON_PARTICIPATING",
72
+ "159": "NON_PARTICIPATING",
73
+ "16": "SUCCESS",
74
  "160": "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": "SUCCESS",
82
+ "168": "NON_PARTICIPATING",
83
+ "169": "NON_PARTICIPATING",
84
  "17": "SUCCESS",
85
+ "170": "SUCCESS",
86
+ "171": "SUCCESS",
87
  "172": "NON_PARTICIPATING",
88
  "173": "NON_PARTICIPATING",
89
  "174": "NON_PARTICIPATING",
 
93
  "178": "NON_PARTICIPATING",
94
  "179": "NON_PARTICIPATING",
95
  "18": "SUCCESS",
96
+ "180": "SUCCESS",
97
  "181": "NON_PARTICIPATING",
98
  "182": "NON_PARTICIPATING",
99
  "183": "NON_PARTICIPATING",
100
+ "184": "SUCCESS",
101
+ "185": "SUCCESS",
102
  "186": "NON_PARTICIPATING",
103
  "187": "NON_PARTICIPATING",
104
  "188": "NON_PARTICIPATING",
105
  "189": "NON_PARTICIPATING",
106
  "19": "SUCCESS",
107
+ "190": "NON_PARTICIPATING",
108
+ "191": "SUCCESS",
109
  "192": "NON_PARTICIPATING",
110
  "193": "NON_PARTICIPATING",
111
  "194": "NON_PARTICIPATING",
 
118
  "20": "SUCCESS",
119
  "200": "NON_PARTICIPATING",
120
  "201": "NON_PARTICIPATING",
121
+ "202": "NON_PARTICIPATING",
122
+ "203": "SUCCESS",
123
  "204": "NON_PARTICIPATING",
124
  "205": "NON_PARTICIPATING",
125
  "206": "NON_PARTICIPATING",
126
  "207": "NON_PARTICIPATING",
127
  "208": "NON_PARTICIPATING",
128
+ "209": "NON_PARTICIPATING",
129
  "21": "SUCCESS",
130
  "210": "NON_PARTICIPATING",
131
  "211": "NON_PARTICIPATING",
 
152
  "230": "NON_PARTICIPATING",
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": "SUCCESS",
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": "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": "SUCCESS",
 
193
  "38": "SUCCESS",
194
  "39": "SUCCESS",
195
  "4": "SUCCESS",
196
+ "40": "SUCCESS",
197
  "41": "SUCCESS",
198
+ "42": "NON_PARTICIPATING",
199
  "43": "SUCCESS",
200
  "44": "SUCCESS",
201
  "45": "SUCCESS",
 
213
  "56": "SUCCESS",
214
  "57": "SUCCESS",
215
  "58": "NON_PARTICIPATING",
216
+ "59": "SUCCESS",
217
  "6": "NON_PARTICIPATING",
218
  "60": "SUCCESS",
219
  "61": "NON_PARTICIPATING",
 
237
  "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
  "8": "SUCCESS",
240
+ "80": "SUCCESS",
241
  "81": "SUCCESS",
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
 
247
  "87": "SUCCESS",
248
  "88": "SUCCESS",
249
  "89": "SUCCESS",
250
+ "9": "SUCCESS",
251
  "90": "SUCCESS",
252
  "91": "SUCCESS",
253
  "92": "SUCCESS",
 
268
  "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
269
  },
270
  "block_list": [
271
+ 5568980,
272
+ 5568981,
273
+ 5568983,
274
+ 5568984,
275
+ 5568986
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 5568918,
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:c168dbe3be5cc8efea844e43eba14dc2f561bcca4639c949bcad5e370a218a10
3
  size 8081782026
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a54ed24188e716091efa8d2c7644edef3bd22232b4b98a010d1a36f466e70221
3
  size 8081782026
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ad3b7c3cff01dad731928c4bb6fe829bab5c1654bdd33d2ae12e8b35bedfb065
3
  size 4040701744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e0bb6a8e948a955f499b1f3826d908182f924141a7c75b181aa0111c5f383772
3
  size 4040701744