kmfoda commited on
Commit
a0c8c5d
·
verified ·
1 Parent(s): 5dd44eb

Run 4. Outer Step 9. Inner Step 0. Peers 84.

Browse files
Files changed (4) hide show
  1. config.json +39 -39
  2. inner_optimizer.pt +1 -1
  3. model.safetensors +1 -1
  4. outer_optimizer.pt +1 -1
config.json CHANGED
@@ -3,8 +3,8 @@
3
  "activation_function": "gelu_new",
4
  "all_reduce_scores": {
5
  "0": "NON_PARTICIPATING",
6
- "1": "SUCCESS",
7
- "10": "FAIL",
8
  "100": "NON_PARTICIPATING",
9
  "101": "SUCCESS",
10
  "102": "NON_PARTICIPATING",
@@ -12,9 +12,9 @@
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",
@@ -26,7 +26,7 @@
26
  "117": "SUCCESS",
27
  "118": "SUCCESS",
28
  "119": "NON_PARTICIPATING",
29
- "12": "SUCCESS",
30
  "120": "NON_PARTICIPATING",
31
  "121": "NON_PARTICIPATING",
32
  "122": "SUCCESS",
@@ -51,26 +51,26 @@
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": "SUCCESS",
58
  "146": "NON_PARTICIPATING",
59
- "147": "SUCCESS",
60
  "148": "SUCCESS",
61
  "149": "SUCCESS",
62
- "15": "FAIL",
63
- "150": "SUCCESS",
64
  "151": "NON_PARTICIPATING",
65
  "152": "NON_PARTICIPATING",
66
- "153": "SUCCESS",
67
- "154": "SUCCESS",
68
  "155": "NON_PARTICIPATING",
69
- "156": "FAIL",
70
  "157": "NON_PARTICIPATING",
71
  "158": "NON_PARTICIPATING",
72
  "159": "SUCCESS",
73
- "16": "SUCCESS",
74
  "160": "SUCCESS",
75
  "161": "NON_PARTICIPATING",
76
  "162": "NON_PARTICIPATING",
@@ -78,7 +78,7 @@
78
  "164": "NON_PARTICIPATING",
79
  "165": "NON_PARTICIPATING",
80
  "166": "NON_PARTICIPATING",
81
- "167": "FAIL",
82
  "168": "NON_PARTICIPATING",
83
  "169": "NON_PARTICIPATING",
84
  "17": "SUCCESS",
@@ -89,8 +89,8 @@
89
  "174": "NON_PARTICIPATING",
90
  "175": "SUCCESS",
91
  "176": "NON_PARTICIPATING",
92
- "177": "FAIL",
93
- "178": "FAIL",
94
  "179": "NON_PARTICIPATING",
95
  "18": "NON_PARTICIPATING",
96
  "180": "NON_PARTICIPATING",
@@ -100,15 +100,15 @@
100
  "184": "NON_PARTICIPATING",
101
  "185": "NON_PARTICIPATING",
102
  "186": "SUCCESS",
103
- "187": "FAIL",
104
  "188": "NON_PARTICIPATING",
105
  "189": "NON_PARTICIPATING",
106
  "19": "NON_PARTICIPATING",
107
  "190": "NON_PARTICIPATING",
108
- "191": "FAIL",
109
  "192": "NON_PARTICIPATING",
110
  "193": "NON_PARTICIPATING",
111
- "194": "NON_PARTICIPATING",
112
  "195": "SUCCESS",
113
  "196": "NON_PARTICIPATING",
114
  "197": "NON_PARTICIPATING",
@@ -117,12 +117,12 @@
117
  "2": "SUCCESS",
118
  "20": "SUCCESS",
119
  "200": "SUCCESS",
120
- "201": "SUCCESS",
121
  "202": "SUCCESS",
122
  "203": "SUCCESS",
123
  "204": "NON_PARTICIPATING",
124
  "205": "NON_PARTICIPATING",
125
- "206": "FAIL",
126
  "207": "NON_PARTICIPATING",
127
  "208": "NON_PARTICIPATING",
128
  "209": "NON_PARTICIPATING",
@@ -133,13 +133,13 @@
133
  "213": "NON_PARTICIPATING",
134
  "214": "NON_PARTICIPATING",
135
  "215": "SUCCESS",
136
- "216": "SUCCESS",
137
  "217": "NON_PARTICIPATING",
138
  "218": "SUCCESS",
139
  "219": "NON_PARTICIPATING",
140
  "22": "SUCCESS",
141
- "220": "FAIL",
142
- "221": "FAIL",
143
  "222": "SUCCESS",
144
  "223": "SUCCESS",
145
  "224": "NON_PARTICIPATING",
@@ -150,7 +150,7 @@
150
  "229": "NON_PARTICIPATING",
151
  "23": "NON_PARTICIPATING",
152
  "230": "NON_PARTICIPATING",
153
- "231": "NON_PARTICIPATING",
154
  "232": "NON_PARTICIPATING",
155
  "233": "NON_PARTICIPATING",
156
  "234": "NON_PARTICIPATING",
@@ -168,7 +168,7 @@
168
  "245": "NON_PARTICIPATING",
169
  "246": "NON_PARTICIPATING",
170
  "247": "SUCCESS",
171
- "248": "NON_PARTICIPATING",
172
  "249": "SUCCESS",
173
  "25": "SUCCESS",
174
  "250": "SUCCESS",
@@ -177,23 +177,23 @@
177
  "253": "NON_PARTICIPATING",
178
  "254": "NON_PARTICIPATING",
179
  "255": "SUCCESS",
180
- "26": "FAIL",
181
  "27": "SUCCESS",
182
  "28": "NON_PARTICIPATING",
183
- "29": "NON_PARTICIPATING",
184
  "3": "NON_PARTICIPATING",
185
  "30": "SUCCESS",
186
  "31": "NON_PARTICIPATING",
187
- "32": "FAIL",
188
  "33": "NON_PARTICIPATING",
189
  "34": "SUCCESS",
190
  "35": "SUCCESS",
191
  "36": "NON_PARTICIPATING",
192
  "37": "SUCCESS",
193
  "38": "NON_PARTICIPATING",
194
- "39": "FAIL",
195
- "4": "FAIL",
196
- "40": "FAIL",
197
  "41": "SUCCESS",
198
  "42": "NON_PARTICIPATING",
199
  "43": "SUCCESS",
@@ -209,12 +209,12 @@
209
  "52": "NON_PARTICIPATING",
210
  "53": "SUCCESS",
211
  "54": "NON_PARTICIPATING",
212
- "55": "SUCCESS",
213
  "56": "SUCCESS",
214
  "57": "NON_PARTICIPATING",
215
  "58": "SUCCESS",
216
  "59": "NON_PARTICIPATING",
217
- "6": "SUCCESS",
218
  "60": "SUCCESS",
219
  "61": "NON_PARTICIPATING",
220
  "62": "SUCCESS",
@@ -242,14 +242,14 @@
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
244
  "84": "SUCCESS",
245
- "85": "SUCCESS",
246
  "86": "NON_PARTICIPATING",
247
  "87": "SUCCESS",
248
- "88": "NON_PARTICIPATING",
249
- "89": "FAIL",
250
  "9": "NON_PARTICIPATING",
251
  "90": "SUCCESS",
252
- "91": "FAIL",
253
  "92": "NON_PARTICIPATING",
254
  "93": "NON_PARTICIPATING",
255
  "94": "NON_PARTICIPATING",
@@ -275,7 +275,7 @@
275
  "initializer_range": 0.02,
276
  "inner_step": 0,
277
  "inner_steps": 0,
278
- "last_allreduce_block": 5499548,
279
  "layer_norm_epsilon": 1e-05,
280
  "model_type": "gpt_optimized",
281
  "n_embd": 1280,
 
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": "SUCCESS",
10
  "102": "NON_PARTICIPATING",
 
12
  "104": "NON_PARTICIPATING",
13
  "105": "NON_PARTICIPATING",
14
  "106": "NON_PARTICIPATING",
15
+ "107": "NON_PARTICIPATING",
16
  "108": "NON_PARTICIPATING",
17
+ "109": "SUCCESS",
18
  "11": "SUCCESS",
19
  "110": "SUCCESS",
20
  "111": "NON_PARTICIPATING",
 
26
  "117": "SUCCESS",
27
  "118": "SUCCESS",
28
  "119": "NON_PARTICIPATING",
29
+ "12": "FAIL",
30
  "120": "NON_PARTICIPATING",
31
  "121": "NON_PARTICIPATING",
32
  "122": "SUCCESS",
 
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": "SUCCESS",
58
  "146": "NON_PARTICIPATING",
59
+ "147": "NON_PARTICIPATING",
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": "NON_PARTICIPATING",
68
  "155": "NON_PARTICIPATING",
69
+ "156": "NON_PARTICIPATING",
70
  "157": "NON_PARTICIPATING",
71
  "158": "NON_PARTICIPATING",
72
  "159": "SUCCESS",
73
+ "16": "NON_PARTICIPATING",
74
  "160": "SUCCESS",
75
  "161": "NON_PARTICIPATING",
76
  "162": "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",
 
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": "NON_PARTICIPATING",
96
  "180": "NON_PARTICIPATING",
 
100
  "184": "NON_PARTICIPATING",
101
  "185": "NON_PARTICIPATING",
102
  "186": "SUCCESS",
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": "NON_PARTICIPATING",
110
  "193": "NON_PARTICIPATING",
111
+ "194": "SUCCESS",
112
  "195": "SUCCESS",
113
  "196": "NON_PARTICIPATING",
114
  "197": "NON_PARTICIPATING",
 
117
  "2": "SUCCESS",
118
  "20": "SUCCESS",
119
  "200": "SUCCESS",
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",
 
133
  "213": "NON_PARTICIPATING",
134
  "214": "NON_PARTICIPATING",
135
  "215": "SUCCESS",
136
+ "216": "NON_PARTICIPATING",
137
  "217": "NON_PARTICIPATING",
138
  "218": "SUCCESS",
139
  "219": "NON_PARTICIPATING",
140
  "22": "SUCCESS",
141
+ "220": "SUCCESS",
142
+ "221": "SUCCESS",
143
  "222": "SUCCESS",
144
  "223": "SUCCESS",
145
  "224": "NON_PARTICIPATING",
 
150
  "229": "NON_PARTICIPATING",
151
  "23": "NON_PARTICIPATING",
152
  "230": "NON_PARTICIPATING",
153
+ "231": "SUCCESS",
154
  "232": "NON_PARTICIPATING",
155
  "233": "NON_PARTICIPATING",
156
  "234": "NON_PARTICIPATING",
 
168
  "245": "NON_PARTICIPATING",
169
  "246": "NON_PARTICIPATING",
170
  "247": "SUCCESS",
171
+ "248": "SUCCESS",
172
  "249": "SUCCESS",
173
  "25": "SUCCESS",
174
  "250": "SUCCESS",
 
177
  "253": "NON_PARTICIPATING",
178
  "254": "NON_PARTICIPATING",
179
  "255": "SUCCESS",
180
+ "26": "SUCCESS",
181
  "27": "SUCCESS",
182
  "28": "NON_PARTICIPATING",
183
+ "29": "FAIL",
184
  "3": "NON_PARTICIPATING",
185
  "30": "SUCCESS",
186
  "31": "NON_PARTICIPATING",
187
+ "32": "SUCCESS",
188
  "33": "NON_PARTICIPATING",
189
  "34": "SUCCESS",
190
  "35": "SUCCESS",
191
  "36": "NON_PARTICIPATING",
192
  "37": "SUCCESS",
193
  "38": "NON_PARTICIPATING",
194
+ "39": "NON_PARTICIPATING",
195
+ "4": "SUCCESS",
196
+ "40": "SUCCESS",
197
  "41": "SUCCESS",
198
  "42": "NON_PARTICIPATING",
199
  "43": "SUCCESS",
 
209
  "52": "NON_PARTICIPATING",
210
  "53": "SUCCESS",
211
  "54": "NON_PARTICIPATING",
212
+ "55": "FAIL",
213
  "56": "SUCCESS",
214
  "57": "NON_PARTICIPATING",
215
  "58": "SUCCESS",
216
  "59": "NON_PARTICIPATING",
217
+ "6": "FAIL",
218
  "60": "SUCCESS",
219
  "61": "NON_PARTICIPATING",
220
  "62": "SUCCESS",
 
242
  "82": "SUCCESS",
243
  "83": "NON_PARTICIPATING",
244
  "84": "SUCCESS",
245
+ "85": "FAIL",
246
  "86": "NON_PARTICIPATING",
247
  "87": "SUCCESS",
248
+ "88": "SUCCESS",
249
+ "89": "SUCCESS",
250
  "9": "NON_PARTICIPATING",
251
  "90": "SUCCESS",
252
+ "91": "SUCCESS",
253
  "92": "NON_PARTICIPATING",
254
  "93": "NON_PARTICIPATING",
255
  "94": "NON_PARTICIPATING",
 
275
  "initializer_range": 0.02,
276
  "inner_step": 0,
277
  "inner_steps": 0,
278
+ "last_allreduce_block": 5501742,
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 sha256:1a606695e0adec9477b43c25b1e1268bab930e40a7fa6c5c73bcb3688e34e5e8
3
  size 8081782026
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b0c5ae9f9907bcaef042e85c359a9bf66210511edc81f26afc0777bae97280f
3
  size 8081782026
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3c1ad854fc14e509aa4a615119d509af3a3d9894ec41b0d8805f1d7f5c626682
3
  size 4040701744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:44a0ac4821bf999e8dba67e03165e62a9aed782e66fef3b101567f778cfdeae7
3
  size 4040701744
outer_optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:207d8838847bb2bb3d0e68733699338cf4c0540a683262b5f75f1bd7077fc90d
3
  size 4040805354
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:58e2cf2581eea7fe4c4932f9118c053583885139150fdb951addcd0b10db2c06
3
  size 4040805354