thieu86 commited on
Commit
dab42a6
·
verified ·
1 Parent(s): b97a1fd

Run 4. Outer Step 60. Inner Step 0.

Browse files
Files changed (3) hide show
  1. config.json +57 -57
  2. inner_optimizer.pt +1 -1
  3. model.safetensors +1 -1
config.json CHANGED
@@ -12,15 +12,15 @@
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": "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",
@@ -28,9 +28,9 @@
28
  "119": "NON_PARTICIPATING",
29
  "12": "SUCCESS",
30
  "120": "NON_PARTICIPATING",
31
- "121": "NON_PARTICIPATING",
32
  "122": "NON_PARTICIPATING",
33
- "123": "SUCCESS",
34
  "124": "SUCCESS",
35
  "125": "NON_PARTICIPATING",
36
  "126": "NON_PARTICIPATING",
@@ -38,33 +38,33 @@
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": "NON_PARTICIPATING",
46
  "135": "NON_PARTICIPATING",
47
- "136": "SUCCESS",
48
  "137": "NON_PARTICIPATING",
49
- "138": "SUCCESS",
50
  "139": "NON_PARTICIPATING",
51
  "14": "SUCCESS",
52
  "140": "NON_PARTICIPATING",
53
  "141": "NON_PARTICIPATING",
54
  "142": "NON_PARTICIPATING",
55
- "143": "SUCCESS",
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": "NON_PARTICIPATING",
67
- "154": "NON_PARTICIPATING",
68
  "155": "NON_PARTICIPATING",
69
  "156": "NON_PARTICIPATING",
70
  "157": "NON_PARTICIPATING",
@@ -84,26 +84,26 @@
84
  "17": "SUCCESS",
85
  "170": "NON_PARTICIPATING",
86
  "171": "NON_PARTICIPATING",
87
- "172": "SUCCESS",
88
- "173": "SUCCESS",
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": "SUCCESS",
96
  "180": "NON_PARTICIPATING",
97
- "181": "NON_PARTICIPATING",
98
  "182": "NON_PARTICIPATING",
99
  "183": "NON_PARTICIPATING",
100
  "184": "NON_PARTICIPATING",
101
  "185": "NON_PARTICIPATING",
102
- "186": "SUCCESS",
103
  "187": "NON_PARTICIPATING",
104
  "188": "NON_PARTICIPATING",
105
- "189": "SUCCESS",
106
- "19": "NON_PARTICIPATING",
107
  "190": "NON_PARTICIPATING",
108
  "191": "NON_PARTICIPATING",
109
  "192": "NON_PARTICIPATING",
@@ -113,37 +113,37 @@
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": "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",
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": "SUCCESS",
136
  "216": "NON_PARTICIPATING",
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": "SUCCESS",
147
  "226": "NON_PARTICIPATING",
148
  "227": "NON_PARTICIPATING",
149
  "228": "NON_PARTICIPATING",
@@ -158,11 +158,11 @@
158
  "236": "NON_PARTICIPATING",
159
  "237": "NON_PARTICIPATING",
160
  "238": "NON_PARTICIPATING",
161
- "239": "SUCCESS",
162
  "24": "SUCCESS",
163
  "240": "NON_PARTICIPATING",
164
- "241": "NON_PARTICIPATING",
165
- "242": "SUCCESS",
166
  "243": "NON_PARTICIPATING",
167
  "244": "NON_PARTICIPATING",
168
  "245": "NON_PARTICIPATING",
@@ -171,24 +171,24 @@
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",
180
  "26": "SUCCESS",
181
  "27": "SUCCESS",
182
  "28": "SUCCESS",
183
  "29": "SUCCESS",
184
  "3": "SUCCESS",
185
- "30": "SUCCESS",
186
- "31": "SUCCESS",
187
  "32": "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",
@@ -200,14 +200,14 @@
200
  "44": "SUCCESS",
201
  "45": "SUCCESS",
202
  "46": "NON_PARTICIPATING",
203
- "47": "NON_PARTICIPATING",
204
  "48": "NON_PARTICIPATING",
205
  "49": "SUCCESS",
206
- "5": "NON_PARTICIPATING",
207
  "50": "SUCCESS",
208
  "51": "SUCCESS",
209
- "52": "SUCCESS",
210
- "53": "NON_PARTICIPATING",
211
  "54": "SUCCESS",
212
  "55": "NON_PARTICIPATING",
213
  "56": "SUCCESS",
@@ -223,8 +223,8 @@
223
  "65": "SUCCESS",
224
  "66": "SUCCESS",
225
  "67": "SUCCESS",
226
- "68": "NON_PARTICIPATING",
227
- "69": "NON_PARTICIPATING",
228
  "7": "SUCCESS",
229
  "70": "NON_PARTICIPATING",
230
  "71": "NON_PARTICIPATING",
@@ -233,7 +233,7 @@
233
  "74": "SUCCESS",
234
  "75": "SUCCESS",
235
  "76": "SUCCESS",
236
- "77": "SUCCESS",
237
  "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
  "8": "SUCCESS",
@@ -242,12 +242,12 @@
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
244
  "84": "SUCCESS",
245
- "85": "SUCCESS",
246
- "86": "SUCCESS",
247
- "87": "SUCCESS",
248
  "88": "SUCCESS",
249
- "89": "SUCCESS",
250
- "9": "NON_PARTICIPATING",
251
  "90": "SUCCESS",
252
  "91": "SUCCESS",
253
  "92": "SUCCESS",
@@ -268,20 +268,20 @@
268
  "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
269
  },
270
  "block_list": [
271
- 5550353,
272
- 5550358,
273
- 5550363,
274
- 5550366,
275
- 5550370
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": 31,
283
  "inner_steps": 0,
284
- "last_allreduce_block": 5546853,
285
  "layer_norm_epsilon": 1e-05,
286
  "model_type": "gpt_optimized",
287
  "n_embd": 1280,
 
12
  "104": "NON_PARTICIPATING",
13
  "105": "NON_PARTICIPATING",
14
  "106": "SUCCESS",
15
+ "107": "SUCCESS",
16
+ "108": "SUCCESS",
17
  "109": "NON_PARTICIPATING",
18
+ "11": "SUCCESS",
19
+ "110": "NON_PARTICIPATING",
20
+ "111": "SUCCESS",
21
  "112": "NON_PARTICIPATING",
22
  "113": "NON_PARTICIPATING",
23
+ "114": "NON_PARTICIPATING",
24
  "115": "NON_PARTICIPATING",
25
  "116": "NON_PARTICIPATING",
26
  "117": "NON_PARTICIPATING",
 
28
  "119": "NON_PARTICIPATING",
29
  "12": "SUCCESS",
30
  "120": "NON_PARTICIPATING",
31
+ "121": "SUCCESS",
32
  "122": "NON_PARTICIPATING",
33
+ "123": "NON_PARTICIPATING",
34
  "124": "SUCCESS",
35
  "125": "NON_PARTICIPATING",
36
  "126": "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": "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",
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": "NON_PARTICIPATING",
62
  "15": "SUCCESS",
63
+ "150": "NON_PARTICIPATING",
64
  "151": "NON_PARTICIPATING",
65
  "152": "NON_PARTICIPATING",
66
  "153": "NON_PARTICIPATING",
67
+ "154": "SUCCESS",
68
  "155": "NON_PARTICIPATING",
69
  "156": "NON_PARTICIPATING",
70
  "157": "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": "NON_PARTICIPATING",
91
  "176": "NON_PARTICIPATING",
92
  "177": "NON_PARTICIPATING",
93
+ "178": "SUCCESS",
94
  "179": "NON_PARTICIPATING",
95
  "18": "SUCCESS",
96
  "180": "NON_PARTICIPATING",
97
+ "181": "SUCCESS",
98
  "182": "NON_PARTICIPATING",
99
  "183": "NON_PARTICIPATING",
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": "SUCCESS",
107
  "190": "NON_PARTICIPATING",
108
  "191": "NON_PARTICIPATING",
109
  "192": "NON_PARTICIPATING",
 
113
  "196": "NON_PARTICIPATING",
114
  "197": "NON_PARTICIPATING",
115
  "198": "NON_PARTICIPATING",
116
+ "199": "SUCCESS",
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",
125
  "206": "NON_PARTICIPATING",
126
  "207": "NON_PARTICIPATING",
127
  "208": "NON_PARTICIPATING",
128
+ "209": "SUCCESS",
129
  "21": "SUCCESS",
130
+ "210": "SUCCESS",
131
  "211": "NON_PARTICIPATING",
132
  "212": "NON_PARTICIPATING",
133
  "213": "NON_PARTICIPATING",
134
  "214": "NON_PARTICIPATING",
135
+ "215": "NON_PARTICIPATING",
136
  "216": "NON_PARTICIPATING",
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": "NON_PARTICIPATING",
147
  "226": "NON_PARTICIPATING",
148
  "227": "NON_PARTICIPATING",
149
  "228": "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": "SUCCESS",
165
+ "242": "NON_PARTICIPATING",
166
  "243": "NON_PARTICIPATING",
167
  "244": "NON_PARTICIPATING",
168
  "245": "NON_PARTICIPATING",
 
171
  "248": "NON_PARTICIPATING",
172
  "249": "NON_PARTICIPATING",
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": "NON_PARTICIPATING",
180
  "26": "SUCCESS",
181
  "27": "SUCCESS",
182
  "28": "SUCCESS",
183
  "29": "SUCCESS",
184
  "3": "SUCCESS",
185
+ "30": "NON_PARTICIPATING",
186
+ "31": "NON_PARTICIPATING",
187
  "32": "SUCCESS",
188
  "33": "NON_PARTICIPATING",
189
+ "34": "SUCCESS",
190
  "35": "SUCCESS",
191
+ "36": "SUCCESS",
192
  "37": "SUCCESS",
193
  "38": "SUCCESS",
194
  "39": "SUCCESS",
 
200
  "44": "SUCCESS",
201
  "45": "SUCCESS",
202
  "46": "NON_PARTICIPATING",
203
+ "47": "SUCCESS",
204
  "48": "NON_PARTICIPATING",
205
  "49": "SUCCESS",
206
+ "5": "SUCCESS",
207
  "50": "SUCCESS",
208
  "51": "SUCCESS",
209
+ "52": "NON_PARTICIPATING",
210
+ "53": "SUCCESS",
211
  "54": "SUCCESS",
212
  "55": "NON_PARTICIPATING",
213
  "56": "SUCCESS",
 
223
  "65": "SUCCESS",
224
  "66": "SUCCESS",
225
  "67": "SUCCESS",
226
+ "68": "SUCCESS",
227
+ "69": "SUCCESS",
228
  "7": "SUCCESS",
229
  "70": "NON_PARTICIPATING",
230
  "71": "NON_PARTICIPATING",
 
233
  "74": "SUCCESS",
234
  "75": "SUCCESS",
235
  "76": "SUCCESS",
236
+ "77": "NON_PARTICIPATING",
237
  "78": "SUCCESS",
238
  "79": "NON_PARTICIPATING",
239
  "8": "SUCCESS",
 
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
244
  "84": "SUCCESS",
245
+ "85": "NON_PARTICIPATING",
246
+ "86": "NON_PARTICIPATING",
247
+ "87": "FAIL",
248
  "88": "SUCCESS",
249
+ "89": "NON_PARTICIPATING",
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
+ 5607765,
272
+ 5607769,
273
+ 5607773,
274
+ 5607777,
275
+ 5607782
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": 5607244,
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:f07f52df18c80e3e81cf2fde9314dbdc099a48d33aaba7c1a7f38b2976eaae83
3
  size 8081782026
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a6f2ca753fdcf14b353d1294109f5b02a02498ee88b2e47c5dd122fa12bba4e
3
  size 8081782026
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:76499c321ea8e131bb68a21aa4f04f5925b85a6c4a53ad774387829b1068f5a4
3
  size 4040701744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:71ebdf993501cb16fb54b328887bc4a06d87255e7dbf734e70458d19fd8f6c13
3
  size 4040701744