kmfoda commited on
Commit
4ebb061
·
verified ·
1 Parent(s): b178c86

Run 6. Outer Step 45. Inner Step 0. Peers 33.

Browse files
Files changed (4) hide show
  1. config.json +21 -21
  2. inner_optimizer.pt +1 -1
  3. model.safetensors +1 -1
  4. outer_optimizer.pt +1 -1
config.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "_name_or_path": "distributed/llama-1b",
3
  "all_reduce_scores": {
4
  "0": "NOT_ALIVE",
5
  "1": "SUCCESS",
@@ -18,7 +18,7 @@
18
  "110": "NOT_ALIVE",
19
  "111": "NOT_ALIVE",
20
  "112": "NOT_ALIVE",
21
- "113": "NOT_ALIVE",
22
  "114": "NOT_ALIVE",
23
  "115": "NOT_ALIVE",
24
  "116": "NOT_ALIVE",
@@ -48,7 +48,7 @@
48
  "138": "SUCCESS",
49
  "139": "NOT_ALIVE",
50
  "14": "NOT_ALIVE",
51
- "140": "SUCCESS",
52
  "141": "NOT_ALIVE",
53
  "142": "NOT_ALIVE",
54
  "143": "NOT_ALIVE",
@@ -74,8 +74,8 @@
74
  "161": "NOT_ALIVE",
75
  "162": "NOT_ALIVE",
76
  "163": "NON_PARTICIPATING",
77
- "164": "NOT_ALIVE",
78
- "165": "NON_PARTICIPATING",
79
  "166": "SUCCESS",
80
  "167": "NOT_ALIVE",
81
  "168": "NOT_ALIVE",
@@ -99,21 +99,21 @@
99
  "184": "NOT_ALIVE",
100
  "185": "NON_PARTICIPATING",
101
  "186": "NOT_ALIVE",
102
- "187": "SUCCESS",
103
  "188": "NOT_ALIVE",
104
  "189": "SUCCESS",
105
  "19": "NOT_ALIVE",
106
  "190": "SUCCESS",
107
- "191": "NON_PARTICIPATING",
108
  "192": "NOT_ALIVE",
109
- "193": "SUCCESS",
110
  "194": "NOT_ALIVE",
111
  "195": "NOT_ALIVE",
112
  "196": "NOT_ALIVE",
113
  "197": "NOT_ALIVE",
114
  "198": "NOT_ALIVE",
115
- "199": "NOT_ALIVE",
116
- "2": "NOT_ALIVE",
117
  "20": "NOT_ALIVE",
118
  "200": "NOT_ALIVE",
119
  "201": "SUCCESS",
@@ -132,22 +132,22 @@
132
  "213": "NOT_ALIVE",
133
  "214": "NON_PARTICIPATING",
134
  "215": "NON_PARTICIPATING",
135
- "216": "NOT_ALIVE",
136
- "217": "SUCCESS",
137
- "218": "NON_PARTICIPATING",
138
  "219": "SUCCESS",
139
  "22": "NOT_ALIVE",
140
  "220": "NOT_ALIVE",
141
  "221": "NOT_ALIVE",
142
  "222": "NOT_ALIVE",
143
  "223": "NOT_ALIVE",
144
- "224": "NOT_ALIVE",
145
  "225": "NOT_ALIVE",
146
  "226": "NON_PARTICIPATING",
147
  "227": "NOT_ALIVE",
148
  "228": "SUCCESS",
149
  "229": "NOT_ALIVE",
150
- "23": "SUCCESS",
151
  "230": "NOT_ALIVE",
152
  "231": "NOT_ALIVE",
153
  "232": "NOT_ALIVE",
@@ -177,12 +177,12 @@
177
  "254": "NOT_ALIVE",
178
  "255": "NOT_ALIVE",
179
  "26": "NON_PARTICIPATING",
180
- "27": "NOT_ALIVE",
181
  "28": "NON_PARTICIPATING",
182
  "29": "NOT_ALIVE",
183
  "3": "SUCCESS",
184
  "30": "NOT_ALIVE",
185
- "31": "SUCCESS",
186
  "32": "SUCCESS",
187
  "33": "NOT_ALIVE",
188
  "34": "NOT_ALIVE",
@@ -200,7 +200,7 @@
200
  "45": "NOT_ALIVE",
201
  "46": "SUCCESS",
202
  "47": "NOT_ALIVE",
203
- "48": "NON_PARTICIPATING",
204
  "49": "NOT_ALIVE",
205
  "5": "NOT_ALIVE",
206
  "50": "NOT_ALIVE",
@@ -209,7 +209,7 @@
209
  "53": "SUCCESS",
210
  "54": "NOT_ALIVE",
211
  "55": "NOT_ALIVE",
212
- "56": "NON_PARTICIPATING",
213
  "57": "NON_PARTICIPATING",
214
  "58": "NOT_ALIVE",
215
  "59": "NOT_ALIVE",
@@ -239,7 +239,7 @@
239
  "80": "NOT_ALIVE",
240
  "81": "NOT_ALIVE",
241
  "82": "NOT_ALIVE",
242
- "83": "NOT_ALIVE",
243
  "84": "NOT_ALIVE",
244
  "85": "NOT_ALIVE",
245
  "86": "NOT_ALIVE",
@@ -271,7 +271,7 @@
271
  "initializer_range": 0.02,
272
  "inner_step": 0,
273
  "intermediate_size": 5632,
274
- "last_allreduce_block": 6158311,
275
  "max_position_embeddings": 2048,
276
  "mlp_bias": false,
277
  "model_type": "llama",
 
1
  {
2
+ "_name_or_path": "asim22/sub38-1",
3
  "all_reduce_scores": {
4
  "0": "NOT_ALIVE",
5
  "1": "SUCCESS",
 
18
  "110": "NOT_ALIVE",
19
  "111": "NOT_ALIVE",
20
  "112": "NOT_ALIVE",
21
+ "113": "SUCCESS",
22
  "114": "NOT_ALIVE",
23
  "115": "NOT_ALIVE",
24
  "116": "NOT_ALIVE",
 
48
  "138": "SUCCESS",
49
  "139": "NOT_ALIVE",
50
  "14": "NOT_ALIVE",
51
+ "140": "NOT_ALIVE",
52
  "141": "NOT_ALIVE",
53
  "142": "NOT_ALIVE",
54
  "143": "NOT_ALIVE",
 
74
  "161": "NOT_ALIVE",
75
  "162": "NOT_ALIVE",
76
  "163": "NON_PARTICIPATING",
77
+ "164": "SUCCESS",
78
+ "165": "NOT_ALIVE",
79
  "166": "SUCCESS",
80
  "167": "NOT_ALIVE",
81
  "168": "NOT_ALIVE",
 
99
  "184": "NOT_ALIVE",
100
  "185": "NON_PARTICIPATING",
101
  "186": "NOT_ALIVE",
102
+ "187": "NOT_ALIVE",
103
  "188": "NOT_ALIVE",
104
  "189": "SUCCESS",
105
  "19": "NOT_ALIVE",
106
  "190": "SUCCESS",
107
+ "191": "SUCCESS",
108
  "192": "NOT_ALIVE",
109
+ "193": "NOT_ALIVE",
110
  "194": "NOT_ALIVE",
111
  "195": "NOT_ALIVE",
112
  "196": "NOT_ALIVE",
113
  "197": "NOT_ALIVE",
114
  "198": "NOT_ALIVE",
115
+ "199": "SUCCESS",
116
+ "2": "SUCCESS",
117
  "20": "NOT_ALIVE",
118
  "200": "NOT_ALIVE",
119
  "201": "SUCCESS",
 
132
  "213": "NOT_ALIVE",
133
  "214": "NON_PARTICIPATING",
134
  "215": "NON_PARTICIPATING",
135
+ "216": "NON_PARTICIPATING",
136
+ "217": "NOT_ALIVE",
137
+ "218": "NOT_ALIVE",
138
  "219": "SUCCESS",
139
  "22": "NOT_ALIVE",
140
  "220": "NOT_ALIVE",
141
  "221": "NOT_ALIVE",
142
  "222": "NOT_ALIVE",
143
  "223": "NOT_ALIVE",
144
+ "224": "NON_PARTICIPATING",
145
  "225": "NOT_ALIVE",
146
  "226": "NON_PARTICIPATING",
147
  "227": "NOT_ALIVE",
148
  "228": "SUCCESS",
149
  "229": "NOT_ALIVE",
150
+ "23": "NOT_ALIVE",
151
  "230": "NOT_ALIVE",
152
  "231": "NOT_ALIVE",
153
  "232": "NOT_ALIVE",
 
177
  "254": "NOT_ALIVE",
178
  "255": "NOT_ALIVE",
179
  "26": "NON_PARTICIPATING",
180
+ "27": "NON_PARTICIPATING",
181
  "28": "NON_PARTICIPATING",
182
  "29": "NOT_ALIVE",
183
  "3": "SUCCESS",
184
  "30": "NOT_ALIVE",
185
+ "31": "NOT_ALIVE",
186
  "32": "SUCCESS",
187
  "33": "NOT_ALIVE",
188
  "34": "NOT_ALIVE",
 
200
  "45": "NOT_ALIVE",
201
  "46": "SUCCESS",
202
  "47": "NOT_ALIVE",
203
+ "48": "NOT_ALIVE",
204
  "49": "NOT_ALIVE",
205
  "5": "NOT_ALIVE",
206
  "50": "NOT_ALIVE",
 
209
  "53": "SUCCESS",
210
  "54": "NOT_ALIVE",
211
  "55": "NOT_ALIVE",
212
+ "56": "NOT_ALIVE",
213
  "57": "NON_PARTICIPATING",
214
  "58": "NOT_ALIVE",
215
  "59": "NOT_ALIVE",
 
239
  "80": "NOT_ALIVE",
240
  "81": "NOT_ALIVE",
241
  "82": "NOT_ALIVE",
242
+ "83": "SUCCESS",
243
  "84": "NOT_ALIVE",
244
  "85": "NOT_ALIVE",
245
  "86": "NOT_ALIVE",
 
271
  "initializer_range": 0.02,
272
  "inner_step": 0,
273
  "intermediate_size": 5632,
274
+ "last_allreduce_block": 6159037,
275
  "max_position_embeddings": 2048,
276
  "mlp_bias": false,
277
  "model_type": "llama",
inner_optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9b21fafa27875327ccfd1b176379181ecbf508ab2fea822adda9d92bc05929f7
3
  size 8800564561
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7e9d67ee00da76368410e2f7304685810339a9387a0d15e5b8951098f44adde0
3
  size 8800564561
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0c28b4321ab01e2005b9e9c6ff098f780c006b656eefe85d604578ff599549a2
3
  size 4400216536
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82f0e253072a2a1b86edf2798f2070e253ca48236b0fd328e182ba092cb115a5
3
  size 4400216536
outer_optimizer.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0120e9680e18b90890fd40b5757c28e3190351ce2339827648b1f9b3d12aaf5a
3
  size 4400252933
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66368eb4702e6c3764502fdf83468251e9978a07abc513a82dd9b4974f13ed2b
3
  size 4400252933