Run 4. Outer Step 8. Inner Step 0.
Browse files- config.json +47 -47
- inner_optimizer.pt +1 -1
- model.safetensors +1 -1
config.json
CHANGED
|
@@ -4,17 +4,17 @@
|
|
| 4 |
"all_reduce_scores": {
|
| 5 |
"0": "NON_PARTICIPATING",
|
| 6 |
"1": "SUCCESS",
|
| 7 |
-
"10": "
|
| 8 |
"100": "NON_PARTICIPATING",
|
| 9 |
"101": "SUCCESS",
|
| 10 |
"102": "NON_PARTICIPATING",
|
| 11 |
"103": "NON_PARTICIPATING",
|
| 12 |
-
"104": "
|
| 13 |
"105": "NON_PARTICIPATING",
|
| 14 |
"106": "NON_PARTICIPATING",
|
| 15 |
"107": "SUCCESS",
|
| 16 |
"108": "NON_PARTICIPATING",
|
| 17 |
-
"109": "
|
| 18 |
"11": "SUCCESS",
|
| 19 |
"110": "SUCCESS",
|
| 20 |
"111": "NON_PARTICIPATING",
|
|
@@ -24,15 +24,15 @@
|
|
| 24 |
"115": "NON_PARTICIPATING",
|
| 25 |
"116": "NON_PARTICIPATING",
|
| 26 |
"117": "SUCCESS",
|
| 27 |
-
"118": "
|
| 28 |
"119": "NON_PARTICIPATING",
|
| 29 |
"12": "SUCCESS",
|
| 30 |
"120": "NON_PARTICIPATING",
|
| 31 |
"121": "NON_PARTICIPATING",
|
| 32 |
"122": "SUCCESS",
|
| 33 |
-
"123": "
|
| 34 |
"124": "NON_PARTICIPATING",
|
| 35 |
-
"125": "
|
| 36 |
"126": "NON_PARTICIPATING",
|
| 37 |
"127": "NON_PARTICIPATING",
|
| 38 |
"128": "NON_PARTICIPATING",
|
|
@@ -43,7 +43,7 @@
|
|
| 43 |
"132": "SUCCESS",
|
| 44 |
"133": "NON_PARTICIPATING",
|
| 45 |
"134": "SUCCESS",
|
| 46 |
-
"135": "
|
| 47 |
"136": "NON_PARTICIPATING",
|
| 48 |
"137": "NON_PARTICIPATING",
|
| 49 |
"138": "NON_PARTICIPATING",
|
|
@@ -54,23 +54,23 @@
|
|
| 54 |
"142": "SUCCESS",
|
| 55 |
"143": "NON_PARTICIPATING",
|
| 56 |
"144": "NON_PARTICIPATING",
|
| 57 |
-
"145": "
|
| 58 |
"146": "NON_PARTICIPATING",
|
| 59 |
-
"147": "
|
| 60 |
"148": "SUCCESS",
|
| 61 |
"149": "SUCCESS",
|
| 62 |
-
"15": "
|
| 63 |
-
"150": "
|
| 64 |
"151": "NON_PARTICIPATING",
|
| 65 |
"152": "NON_PARTICIPATING",
|
| 66 |
-
"153": "
|
| 67 |
"154": "SUCCESS",
|
| 68 |
"155": "NON_PARTICIPATING",
|
| 69 |
-
"156": "
|
| 70 |
"157": "NON_PARTICIPATING",
|
| 71 |
"158": "NON_PARTICIPATING",
|
| 72 |
-
"159": "
|
| 73 |
-
"16": "
|
| 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": "
|
| 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": "
|
| 93 |
-
"178": "
|
| 94 |
"179": "NON_PARTICIPATING",
|
| 95 |
"18": "NON_PARTICIPATING",
|
| 96 |
"180": "NON_PARTICIPATING",
|
|
@@ -100,13 +100,13 @@
|
|
| 100 |
"184": "NON_PARTICIPATING",
|
| 101 |
"185": "NON_PARTICIPATING",
|
| 102 |
"186": "SUCCESS",
|
| 103 |
-
"187": "
|
| 104 |
"188": "NON_PARTICIPATING",
|
| 105 |
"189": "NON_PARTICIPATING",
|
| 106 |
"19": "NON_PARTICIPATING",
|
| 107 |
"190": "NON_PARTICIPATING",
|
| 108 |
-
"191": "
|
| 109 |
-
"192": "
|
| 110 |
"193": "NON_PARTICIPATING",
|
| 111 |
"194": "NON_PARTICIPATING",
|
| 112 |
"195": "SUCCESS",
|
|
@@ -117,12 +117,12 @@
|
|
| 117 |
"2": "SUCCESS",
|
| 118 |
"20": "SUCCESS",
|
| 119 |
"200": "SUCCESS",
|
| 120 |
-
"201": "
|
| 121 |
-
"202": "
|
| 122 |
"203": "SUCCESS",
|
| 123 |
"204": "NON_PARTICIPATING",
|
| 124 |
"205": "NON_PARTICIPATING",
|
| 125 |
-
"206": "
|
| 126 |
"207": "NON_PARTICIPATING",
|
| 127 |
"208": "NON_PARTICIPATING",
|
| 128 |
"209": "NON_PARTICIPATING",
|
|
@@ -132,23 +132,23 @@
|
|
| 132 |
"212": "NON_PARTICIPATING",
|
| 133 |
"213": "NON_PARTICIPATING",
|
| 134 |
"214": "NON_PARTICIPATING",
|
| 135 |
-
"215": "
|
| 136 |
"216": "SUCCESS",
|
| 137 |
"217": "NON_PARTICIPATING",
|
| 138 |
"218": "SUCCESS",
|
| 139 |
"219": "NON_PARTICIPATING",
|
| 140 |
"22": "SUCCESS",
|
| 141 |
-
"220": "
|
| 142 |
-
"221": "
|
| 143 |
"222": "SUCCESS",
|
| 144 |
"223": "SUCCESS",
|
| 145 |
"224": "NON_PARTICIPATING",
|
| 146 |
"225": "SUCCESS",
|
| 147 |
-
"226": "
|
| 148 |
"227": "NON_PARTICIPATING",
|
| 149 |
"228": "NON_PARTICIPATING",
|
| 150 |
"229": "NON_PARTICIPATING",
|
| 151 |
-
"23": "
|
| 152 |
"230": "NON_PARTICIPATING",
|
| 153 |
"231": "NON_PARTICIPATING",
|
| 154 |
"232": "NON_PARTICIPATING",
|
|
@@ -177,29 +177,29 @@
|
|
| 177 |
"253": "NON_PARTICIPATING",
|
| 178 |
"254": "NON_PARTICIPATING",
|
| 179 |
"255": "SUCCESS",
|
| 180 |
-
"26": "
|
| 181 |
"27": "SUCCESS",
|
| 182 |
"28": "NON_PARTICIPATING",
|
| 183 |
"29": "NON_PARTICIPATING",
|
| 184 |
"3": "NON_PARTICIPATING",
|
| 185 |
"30": "SUCCESS",
|
| 186 |
"31": "NON_PARTICIPATING",
|
| 187 |
-
"32": "
|
| 188 |
"33": "NON_PARTICIPATING",
|
| 189 |
"34": "SUCCESS",
|
| 190 |
-
"35": "
|
| 191 |
"36": "NON_PARTICIPATING",
|
| 192 |
"37": "SUCCESS",
|
| 193 |
"38": "NON_PARTICIPATING",
|
| 194 |
-
"39": "
|
| 195 |
-
"4": "
|
| 196 |
-
"40": "
|
| 197 |
"41": "SUCCESS",
|
| 198 |
"42": "NON_PARTICIPATING",
|
| 199 |
"43": "SUCCESS",
|
| 200 |
"44": "SUCCESS",
|
| 201 |
"45": "SUCCESS",
|
| 202 |
-
"46": "
|
| 203 |
"47": "NON_PARTICIPATING",
|
| 204 |
"48": "NON_PARTICIPATING",
|
| 205 |
"49": "SUCCESS",
|
|
@@ -245,11 +245,11 @@
|
|
| 245 |
"85": "SUCCESS",
|
| 246 |
"86": "NON_PARTICIPATING",
|
| 247 |
"87": "SUCCESS",
|
| 248 |
-
"88": "
|
| 249 |
-
"89": "
|
| 250 |
"9": "NON_PARTICIPATING",
|
| 251 |
-
"90": "
|
| 252 |
-
"91": "
|
| 253 |
"92": "NON_PARTICIPATING",
|
| 254 |
"93": "NON_PARTICIPATING",
|
| 255 |
"94": "NON_PARTICIPATING",
|
|
@@ -268,20 +268,20 @@
|
|
| 268 |
"AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
|
| 269 |
},
|
| 270 |
"block_list": [
|
| 271 |
-
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
|
| 275 |
-
|
| 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":
|
| 283 |
"inner_steps": 0,
|
| 284 |
-
"last_allreduce_block":
|
| 285 |
"layer_norm_epsilon": 1e-05,
|
| 286 |
"model_type": "gpt_optimized",
|
| 287 |
"n_embd": 1280,
|
|
|
|
| 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",
|
| 11 |
"103": "NON_PARTICIPATING",
|
| 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",
|
|
|
|
| 24 |
"115": "NON_PARTICIPATING",
|
| 25 |
"116": "NON_PARTICIPATING",
|
| 26 |
"117": "SUCCESS",
|
| 27 |
+
"118": "SUCCESS",
|
| 28 |
"119": "NON_PARTICIPATING",
|
| 29 |
"12": "SUCCESS",
|
| 30 |
"120": "NON_PARTICIPATING",
|
| 31 |
"121": "NON_PARTICIPATING",
|
| 32 |
"122": "SUCCESS",
|
| 33 |
+
"123": "SUCCESS",
|
| 34 |
"124": "NON_PARTICIPATING",
|
| 35 |
+
"125": "NON_PARTICIPATING",
|
| 36 |
"126": "NON_PARTICIPATING",
|
| 37 |
"127": "NON_PARTICIPATING",
|
| 38 |
"128": "NON_PARTICIPATING",
|
|
|
|
| 43 |
"132": "SUCCESS",
|
| 44 |
"133": "NON_PARTICIPATING",
|
| 45 |
"134": "SUCCESS",
|
| 46 |
+
"135": "NON_PARTICIPATING",
|
| 47 |
"136": "NON_PARTICIPATING",
|
| 48 |
"137": "NON_PARTICIPATING",
|
| 49 |
"138": "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 |
"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 |
"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 |
"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",
|
|
|
|
| 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",
|
|
|
|
| 132 |
"212": "NON_PARTICIPATING",
|
| 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",
|
| 146 |
"225": "SUCCESS",
|
| 147 |
+
"226": "SUCCESS",
|
| 148 |
"227": "NON_PARTICIPATING",
|
| 149 |
"228": "NON_PARTICIPATING",
|
| 150 |
"229": "NON_PARTICIPATING",
|
| 151 |
+
"23": "NON_PARTICIPATING",
|
| 152 |
"230": "NON_PARTICIPATING",
|
| 153 |
"231": "NON_PARTICIPATING",
|
| 154 |
"232": "NON_PARTICIPATING",
|
|
|
|
| 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",
|
| 200 |
"44": "SUCCESS",
|
| 201 |
"45": "SUCCESS",
|
| 202 |
+
"46": "SUCCESS",
|
| 203 |
"47": "NON_PARTICIPATING",
|
| 204 |
"48": "NON_PARTICIPATING",
|
| 205 |
"49": "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",
|
|
|
|
| 268 |
"AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
|
| 269 |
},
|
| 270 |
"block_list": [
|
| 271 |
+
5499908,
|
| 272 |
+
5499912,
|
| 273 |
+
5499916,
|
| 274 |
+
5499920,
|
| 275 |
+
5499924
|
| 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": 5499548,
|
| 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:
|
| 3 |
size 8081782026
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1a606695e0adec9477b43c25b1e1268bab930e40a7fa6c5c73bcb3688e34e5e8
|
| 3 |
size 8081782026
|
model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4040701744
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3c1ad854fc14e509aa4a615119d509af3a3d9894ec41b0d8805f1d7f5c626682
|
| 3 |
size 4040701744
|