Run 5. Outer Step 77. Inner Step 1.
Browse files- config.json +35 -37
- inner_optimizer.pt +1 -1
- model.safetensors +1 -1
config.json
CHANGED
|
@@ -2,8 +2,8 @@
|
|
| 2 |
"_name_or_path": "distributed/llama-1b",
|
| 3 |
"all_reduce_scores": {
|
| 4 |
"0": "NOT_ALIVE",
|
| 5 |
-
"1": "
|
| 6 |
-
"10": "
|
| 7 |
"100": "NOT_ALIVE",
|
| 8 |
"101": "NOT_ALIVE",
|
| 9 |
"102": "NON_PARTICIPATING",
|
|
@@ -22,20 +22,20 @@
|
|
| 22 |
"114": "NON_PARTICIPATING",
|
| 23 |
"115": "NOT_ALIVE",
|
| 24 |
"116": "NOT_ALIVE",
|
| 25 |
-
"117": "
|
| 26 |
-
"118": "
|
| 27 |
"119": "NOT_ALIVE",
|
| 28 |
"12": "NON_PARTICIPATING",
|
| 29 |
"120": "NON_PARTICIPATING",
|
| 30 |
"121": "NOT_ALIVE",
|
| 31 |
"122": "NOT_ALIVE",
|
| 32 |
"123": "NOT_ALIVE",
|
| 33 |
-
"124": "
|
| 34 |
"125": "NON_PARTICIPATING",
|
| 35 |
"126": "NON_PARTICIPATING",
|
| 36 |
"127": "NOT_ALIVE",
|
| 37 |
"128": "NOT_ALIVE",
|
| 38 |
-
"129": "
|
| 39 |
"13": "NOT_ALIVE",
|
| 40 |
"130": "NOT_ALIVE",
|
| 41 |
"131": "NOT_ALIVE",
|
|
@@ -45,7 +45,7 @@
|
|
| 45 |
"135": "NON_PARTICIPATING",
|
| 46 |
"136": "NOT_ALIVE",
|
| 47 |
"137": "NON_PARTICIPATING",
|
| 48 |
-
"138": "
|
| 49 |
"139": "NON_PARTICIPATING",
|
| 50 |
"14": "NON_PARTICIPATING",
|
| 51 |
"140": "NOT_ALIVE",
|
|
@@ -70,7 +70,7 @@
|
|
| 70 |
"158": "NOT_ALIVE",
|
| 71 |
"159": "NOT_ALIVE",
|
| 72 |
"16": "NON_PARTICIPATING",
|
| 73 |
-
"160": "
|
| 74 |
"161": "NOT_ALIVE",
|
| 75 |
"162": "NON_PARTICIPATING",
|
| 76 |
"163": "NOT_ALIVE",
|
|
@@ -96,7 +96,7 @@
|
|
| 96 |
"181": "NON_PARTICIPATING",
|
| 97 |
"182": "NON_PARTICIPATING",
|
| 98 |
"183": "SUCCESS",
|
| 99 |
-
"184": "
|
| 100 |
"185": "NON_PARTICIPATING",
|
| 101 |
"186": "NON_PARTICIPATING",
|
| 102 |
"187": "SUCCESS",
|
|
@@ -117,7 +117,7 @@
|
|
| 117 |
"20": "NOT_ALIVE",
|
| 118 |
"200": "NON_PARTICIPATING",
|
| 119 |
"201": "NOT_ALIVE",
|
| 120 |
-
"202": "
|
| 121 |
"203": "NOT_ALIVE",
|
| 122 |
"204": "NON_PARTICIPATING",
|
| 123 |
"205": "NOT_ALIVE",
|
|
@@ -136,7 +136,7 @@
|
|
| 136 |
"217": "NOT_ALIVE",
|
| 137 |
"218": "NON_PARTICIPATING",
|
| 138 |
"219": "NON_PARTICIPATING",
|
| 139 |
-
"22": "
|
| 140 |
"220": "NON_PARTICIPATING",
|
| 141 |
"221": "NON_PARTICIPATING",
|
| 142 |
"222": "NOT_ALIVE",
|
|
@@ -144,7 +144,7 @@
|
|
| 144 |
"224": "NOT_ALIVE",
|
| 145 |
"225": "NOT_ALIVE",
|
| 146 |
"226": "NOT_ALIVE",
|
| 147 |
-
"227": "
|
| 148 |
"228": "NOT_ALIVE",
|
| 149 |
"229": "NOT_ALIVE",
|
| 150 |
"23": "NOT_ALIVE",
|
|
@@ -156,7 +156,7 @@
|
|
| 156 |
"235": "NOT_ALIVE",
|
| 157 |
"236": "NOT_ALIVE",
|
| 158 |
"237": "NOT_ALIVE",
|
| 159 |
-
"238": "
|
| 160 |
"239": "NOT_ALIVE",
|
| 161 |
"24": "NOT_ALIVE",
|
| 162 |
"240": "NOT_ALIVE",
|
|
@@ -176,7 +176,7 @@
|
|
| 176 |
"253": "NOT_ALIVE",
|
| 177 |
"254": "NOT_ALIVE",
|
| 178 |
"255": "NOT_ALIVE",
|
| 179 |
-
"26": "
|
| 180 |
"27": "SUCCESS",
|
| 181 |
"28": "SUCCESS",
|
| 182 |
"29": "NOT_ALIVE",
|
|
@@ -191,18 +191,18 @@
|
|
| 191 |
"37": "NON_PARTICIPATING",
|
| 192 |
"38": "SUCCESS",
|
| 193 |
"39": "NOT_ALIVE",
|
| 194 |
-
"4": "
|
| 195 |
"40": "NON_PARTICIPATING",
|
| 196 |
"41": "SUCCESS",
|
| 197 |
"42": "SUCCESS",
|
| 198 |
-
"43": "
|
| 199 |
"44": "NOT_ALIVE",
|
| 200 |
"45": "NON_PARTICIPATING",
|
| 201 |
"46": "NOT_ALIVE",
|
| 202 |
"47": "NON_PARTICIPATING",
|
| 203 |
-
"48": "
|
| 204 |
"49": "NOT_ALIVE",
|
| 205 |
-
"5": "
|
| 206 |
"50": "NOT_ALIVE",
|
| 207 |
"51": "NON_PARTICIPATING",
|
| 208 |
"52": "NOT_ALIVE",
|
|
@@ -210,12 +210,12 @@
|
|
| 210 |
"54": "SUCCESS",
|
| 211 |
"55": "NON_PARTICIPATING",
|
| 212 |
"56": "SUCCESS",
|
| 213 |
-
"57": "
|
| 214 |
"58": "NON_PARTICIPATING",
|
| 215 |
-
"59": "
|
| 216 |
"6": "NOT_ALIVE",
|
| 217 |
"60": "NON_PARTICIPATING",
|
| 218 |
-
"61": "
|
| 219 |
"62": "NON_PARTICIPATING",
|
| 220 |
"63": "NON_PARTICIPATING",
|
| 221 |
"64": "NON_PARTICIPATING",
|
|
@@ -223,14 +223,14 @@
|
|
| 223 |
"66": "NON_PARTICIPATING",
|
| 224 |
"67": "NOT_ALIVE",
|
| 225 |
"68": "NON_PARTICIPATING",
|
| 226 |
-
"69": "
|
| 227 |
"7": "SUCCESS",
|
| 228 |
-
"70": "
|
| 229 |
"71": "NON_PARTICIPATING",
|
| 230 |
"72": "NON_PARTICIPATING",
|
| 231 |
"73": "SUCCESS",
|
| 232 |
"74": "NON_PARTICIPATING",
|
| 233 |
-
"75": "
|
| 234 |
"76": "NON_PARTICIPATING",
|
| 235 |
"77": "NOT_ALIVE",
|
| 236 |
"78": "NOT_ALIVE",
|
|
@@ -241,22 +241,22 @@
|
|
| 241 |
"82": "NON_PARTICIPATING",
|
| 242 |
"83": "NOT_ALIVE",
|
| 243 |
"84": "SUCCESS",
|
| 244 |
-
"85": "
|
| 245 |
"86": "SUCCESS",
|
| 246 |
-
"87": "
|
| 247 |
"88": "NON_PARTICIPATING",
|
| 248 |
"89": "NOT_ALIVE",
|
| 249 |
-
"9": "
|
| 250 |
"90": "SUCCESS",
|
| 251 |
"91": "NOT_ALIVE",
|
| 252 |
-
"92": "
|
| 253 |
"93": "NON_PARTICIPATING",
|
| 254 |
"94": "NON_PARTICIPATING",
|
| 255 |
"95": "NON_PARTICIPATING",
|
| 256 |
-
"96": "
|
| 257 |
"97": "SUCCESS",
|
| 258 |
-
"98": "
|
| 259 |
-
"99": "
|
| 260 |
},
|
| 261 |
"architectures": [
|
| 262 |
"LlamaForCausalLM"
|
|
@@ -264,19 +264,17 @@
|
|
| 264 |
"attention_bias": false,
|
| 265 |
"attention_dropout": 0.0,
|
| 266 |
"block_list": [
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
5826686,
|
| 270 |
-
5826690
|
| 271 |
],
|
| 272 |
"bos_token_id": 1,
|
| 273 |
"eos_token_id": 2,
|
| 274 |
"hidden_act": "silu",
|
| 275 |
"hidden_size": 2048,
|
| 276 |
"initializer_range": 0.02,
|
| 277 |
-
"inner_step":
|
| 278 |
"intermediate_size": 5632,
|
| 279 |
-
"last_allreduce_block":
|
| 280 |
"max_position_embeddings": 2048,
|
| 281 |
"mlp_bias": false,
|
| 282 |
"model_type": "llama",
|
|
|
|
| 2 |
"_name_or_path": "distributed/llama-1b",
|
| 3 |
"all_reduce_scores": {
|
| 4 |
"0": "NOT_ALIVE",
|
| 5 |
+
"1": "NOT_ALIVE",
|
| 6 |
+
"10": "NON_PARTICIPATING",
|
| 7 |
"100": "NOT_ALIVE",
|
| 8 |
"101": "NOT_ALIVE",
|
| 9 |
"102": "NON_PARTICIPATING",
|
|
|
|
| 22 |
"114": "NON_PARTICIPATING",
|
| 23 |
"115": "NOT_ALIVE",
|
| 24 |
"116": "NOT_ALIVE",
|
| 25 |
+
"117": "SUCCESS",
|
| 26 |
+
"118": "NOT_ALIVE",
|
| 27 |
"119": "NOT_ALIVE",
|
| 28 |
"12": "NON_PARTICIPATING",
|
| 29 |
"120": "NON_PARTICIPATING",
|
| 30 |
"121": "NOT_ALIVE",
|
| 31 |
"122": "NOT_ALIVE",
|
| 32 |
"123": "NOT_ALIVE",
|
| 33 |
+
"124": "NON_PARTICIPATING",
|
| 34 |
"125": "NON_PARTICIPATING",
|
| 35 |
"126": "NON_PARTICIPATING",
|
| 36 |
"127": "NOT_ALIVE",
|
| 37 |
"128": "NOT_ALIVE",
|
| 38 |
+
"129": "NON_PARTICIPATING",
|
| 39 |
"13": "NOT_ALIVE",
|
| 40 |
"130": "NOT_ALIVE",
|
| 41 |
"131": "NOT_ALIVE",
|
|
|
|
| 45 |
"135": "NON_PARTICIPATING",
|
| 46 |
"136": "NOT_ALIVE",
|
| 47 |
"137": "NON_PARTICIPATING",
|
| 48 |
+
"138": "SUCCESS",
|
| 49 |
"139": "NON_PARTICIPATING",
|
| 50 |
"14": "NON_PARTICIPATING",
|
| 51 |
"140": "NOT_ALIVE",
|
|
|
|
| 70 |
"158": "NOT_ALIVE",
|
| 71 |
"159": "NOT_ALIVE",
|
| 72 |
"16": "NON_PARTICIPATING",
|
| 73 |
+
"160": "NOT_ALIVE",
|
| 74 |
"161": "NOT_ALIVE",
|
| 75 |
"162": "NON_PARTICIPATING",
|
| 76 |
"163": "NOT_ALIVE",
|
|
|
|
| 96 |
"181": "NON_PARTICIPATING",
|
| 97 |
"182": "NON_PARTICIPATING",
|
| 98 |
"183": "SUCCESS",
|
| 99 |
+
"184": "NOT_ALIVE",
|
| 100 |
"185": "NON_PARTICIPATING",
|
| 101 |
"186": "NON_PARTICIPATING",
|
| 102 |
"187": "SUCCESS",
|
|
|
|
| 117 |
"20": "NOT_ALIVE",
|
| 118 |
"200": "NON_PARTICIPATING",
|
| 119 |
"201": "NOT_ALIVE",
|
| 120 |
+
"202": "NOT_ALIVE",
|
| 121 |
"203": "NOT_ALIVE",
|
| 122 |
"204": "NON_PARTICIPATING",
|
| 123 |
"205": "NOT_ALIVE",
|
|
|
|
| 136 |
"217": "NOT_ALIVE",
|
| 137 |
"218": "NON_PARTICIPATING",
|
| 138 |
"219": "NON_PARTICIPATING",
|
| 139 |
+
"22": "NOT_ALIVE",
|
| 140 |
"220": "NON_PARTICIPATING",
|
| 141 |
"221": "NON_PARTICIPATING",
|
| 142 |
"222": "NOT_ALIVE",
|
|
|
|
| 144 |
"224": "NOT_ALIVE",
|
| 145 |
"225": "NOT_ALIVE",
|
| 146 |
"226": "NOT_ALIVE",
|
| 147 |
+
"227": "NOT_ALIVE",
|
| 148 |
"228": "NOT_ALIVE",
|
| 149 |
"229": "NOT_ALIVE",
|
| 150 |
"23": "NOT_ALIVE",
|
|
|
|
| 156 |
"235": "NOT_ALIVE",
|
| 157 |
"236": "NOT_ALIVE",
|
| 158 |
"237": "NOT_ALIVE",
|
| 159 |
+
"238": "NOT_ALIVE",
|
| 160 |
"239": "NOT_ALIVE",
|
| 161 |
"24": "NOT_ALIVE",
|
| 162 |
"240": "NOT_ALIVE",
|
|
|
|
| 176 |
"253": "NOT_ALIVE",
|
| 177 |
"254": "NOT_ALIVE",
|
| 178 |
"255": "NOT_ALIVE",
|
| 179 |
+
"26": "NON_PARTICIPATING",
|
| 180 |
"27": "SUCCESS",
|
| 181 |
"28": "SUCCESS",
|
| 182 |
"29": "NOT_ALIVE",
|
|
|
|
| 191 |
"37": "NON_PARTICIPATING",
|
| 192 |
"38": "SUCCESS",
|
| 193 |
"39": "NOT_ALIVE",
|
| 194 |
+
"4": "NOT_ALIVE",
|
| 195 |
"40": "NON_PARTICIPATING",
|
| 196 |
"41": "SUCCESS",
|
| 197 |
"42": "SUCCESS",
|
| 198 |
+
"43": "NOT_ALIVE",
|
| 199 |
"44": "NOT_ALIVE",
|
| 200 |
"45": "NON_PARTICIPATING",
|
| 201 |
"46": "NOT_ALIVE",
|
| 202 |
"47": "NON_PARTICIPATING",
|
| 203 |
+
"48": "SUCCESS",
|
| 204 |
"49": "NOT_ALIVE",
|
| 205 |
+
"5": "NON_PARTICIPATING",
|
| 206 |
"50": "NOT_ALIVE",
|
| 207 |
"51": "NON_PARTICIPATING",
|
| 208 |
"52": "NOT_ALIVE",
|
|
|
|
| 210 |
"54": "SUCCESS",
|
| 211 |
"55": "NON_PARTICIPATING",
|
| 212 |
"56": "SUCCESS",
|
| 213 |
+
"57": "NOT_ALIVE",
|
| 214 |
"58": "NON_PARTICIPATING",
|
| 215 |
+
"59": "NON_PARTICIPATING",
|
| 216 |
"6": "NOT_ALIVE",
|
| 217 |
"60": "NON_PARTICIPATING",
|
| 218 |
+
"61": "NON_PARTICIPATING",
|
| 219 |
"62": "NON_PARTICIPATING",
|
| 220 |
"63": "NON_PARTICIPATING",
|
| 221 |
"64": "NON_PARTICIPATING",
|
|
|
|
| 223 |
"66": "NON_PARTICIPATING",
|
| 224 |
"67": "NOT_ALIVE",
|
| 225 |
"68": "NON_PARTICIPATING",
|
| 226 |
+
"69": "NON_PARTICIPATING",
|
| 227 |
"7": "SUCCESS",
|
| 228 |
+
"70": "NOT_ALIVE",
|
| 229 |
"71": "NON_PARTICIPATING",
|
| 230 |
"72": "NON_PARTICIPATING",
|
| 231 |
"73": "SUCCESS",
|
| 232 |
"74": "NON_PARTICIPATING",
|
| 233 |
+
"75": "SUCCESS",
|
| 234 |
"76": "NON_PARTICIPATING",
|
| 235 |
"77": "NOT_ALIVE",
|
| 236 |
"78": "NOT_ALIVE",
|
|
|
|
| 241 |
"82": "NON_PARTICIPATING",
|
| 242 |
"83": "NOT_ALIVE",
|
| 243 |
"84": "SUCCESS",
|
| 244 |
+
"85": "NON_PARTICIPATING",
|
| 245 |
"86": "SUCCESS",
|
| 246 |
+
"87": "NOT_ALIVE",
|
| 247 |
"88": "NON_PARTICIPATING",
|
| 248 |
"89": "NOT_ALIVE",
|
| 249 |
+
"9": "NON_PARTICIPATING",
|
| 250 |
"90": "SUCCESS",
|
| 251 |
"91": "NOT_ALIVE",
|
| 252 |
+
"92": "SUCCESS",
|
| 253 |
"93": "NON_PARTICIPATING",
|
| 254 |
"94": "NON_PARTICIPATING",
|
| 255 |
"95": "NON_PARTICIPATING",
|
| 256 |
+
"96": "NOT_ALIVE",
|
| 257 |
"97": "SUCCESS",
|
| 258 |
+
"98": "SUCCESS",
|
| 259 |
+
"99": "NOT_ALIVE"
|
| 260 |
},
|
| 261 |
"architectures": [
|
| 262 |
"LlamaForCausalLM"
|
|
|
|
| 264 |
"attention_bias": false,
|
| 265 |
"attention_dropout": 0.0,
|
| 266 |
"block_list": [
|
| 267 |
+
5826737,
|
| 268 |
+
5826741
|
|
|
|
|
|
|
| 269 |
],
|
| 270 |
"bos_token_id": 1,
|
| 271 |
"eos_token_id": 2,
|
| 272 |
"hidden_act": "silu",
|
| 273 |
"hidden_size": 2048,
|
| 274 |
"initializer_range": 0.02,
|
| 275 |
+
"inner_step": 1,
|
| 276 |
"intermediate_size": 5632,
|
| 277 |
+
"last_allreduce_block": 5826608,
|
| 278 |
"max_position_embeddings": 2048,
|
| 279 |
"mlp_bias": false,
|
| 280 |
"model_type": "llama",
|
inner_optimizer.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 8800564561
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2659c7916c5dd48cfbdcef1ddbbfc64a7eca31790e21550c1c68d078310e7388
|
| 3 |
size 8800564561
|
model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4400216536
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:806a72856741e15745a5c376b48470d60379a72cb451eef1c7e322efe07dd841
|
| 3 |
size 4400216536
|