Run 7. Outer Step 133. Inner Step 0. Peers 33.

config.json

@@ -4,7 +4,7 @@
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
-    "100": "NON_PARTICIPATING",
+    "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
@@ -16,13 +16,13 @@
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
-    "111": "NOT_ALIVE",
-    "112": "NOT_ALIVE",
+    "111": "SUCCESS",
+    "112": "NON_PARTICIPATING",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
-    "117": "SUCCESS",
+    "117": "NON_PARTICIPATING",
     "118": "SUCCESS",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
@@ -40,7 +40,7 @@
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "SUCCESS",
+    "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
@@ -51,8 +51,8 @@
     "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
-    "143": "SUCCESS",
-    "144": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
+    "144": "SUCCESS",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
@@ -63,8 +63,8 @@
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
-    "154": "SUCCESS",
-    "155": "SUCCESS",
+    "154": "NON_PARTICIPATING",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
@@ -75,31 +75,31 @@
     "162": "NOT_ALIVE",
     "163": "SUCCESS",
     "164": "NON_PARTICIPATING",
-    "165": "NOT_ALIVE",
+    "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NON_PARTICIPATING",
-    "173": "NON_PARTICIPATING",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
-    "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
+    "18": "SUCCESS",
+    "180": "NOT_ALIVE",
     "181": "SUCCESS",
     "182": "NOT_ALIVE",
-    "183": "NOT_ALIVE",
+    "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
     "186": "SUCCESS",
-    "187": "NOT_ALIVE",
+    "187": "SUCCESS",
     "188": "NON_PARTICIPATING",
     "189": "SUCCESS",
     "19": "NOT_ALIVE",
@@ -111,7 +111,7 @@
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NOT_ALIVE",
+    "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
@@ -125,23 +125,23 @@
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "NOT_ALIVE",
-    "21": "NOT_ALIVE",
+    "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
     "211": "SUCCESS",
-    "212": "NOT_ALIVE",
+    "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
-    "214": "NON_PARTICIPATING",
+    "214": "SUCCESS",
     "215": "NOT_ALIVE",
-    "216": "NON_PARTICIPATING",
+    "216": "SUCCESS",
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
-    "219": "NON_PARTICIPATING",
+    "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
     "222": "SUCCESS",
     "223": "NOT_ALIVE",
-    "224": "NOT_ALIVE",
+    "224": "NON_PARTICIPATING",
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
@@ -152,18 +152,18 @@
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
-    "234": "SUCCESS",
+    "234": "NON_PARTICIPATING",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
-    "239": "NOT_ALIVE",
+    "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
-    "242": "SUCCESS",
+    "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
-    "244": "NOT_ALIVE",
+    "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
     "246": "SUCCESS",
     "247": "NON_PARTICIPATING",
@@ -176,26 +176,26 @@
     "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "NOT_ALIVE",
     "27": "SUCCESS",
-    "28": "NOT_ALIVE",
+    "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "NOT_ALIVE",
+    "3": "SUCCESS",
     "30": "NOT_ALIVE",
     "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
-    "36": "SUCCESS",
+    "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
-    "41": "NOT_ALIVE",
+    "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
-    "43": "NOT_ALIVE",
+    "43": "NON_PARTICIPATING",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
@@ -215,10 +215,10 @@
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "61": "NOT_ALIVE",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
-    "64": "SUCCESS",
+    "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
@@ -233,10 +233,10 @@
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
-    "78": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
@@ -245,7 +245,7 @@
     "86": "NON_PARTICIPATING",
     "87": "NOT_ALIVE",
     "88": "NOT_ALIVE",
-    "89": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6316594,
+  "last_allreduce_block": 6317391,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42d8b741c877d2df4328118121fd8082132bbbd381be9a8d76889c8adbaa8efc
+oid sha256:21880feada2ddfc9c41ddc1a25a3cc608ad25642ee3ae0b9587b6e97e83588a6
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f994820b640d43d9c085e88abfb52227a0100795b22b18efa19f5b91068d5d28
+oid sha256:910e7c869b285bbc508e2935c57b7b9903642c015d24cafd5d976094ecea39ec
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f528447b4c99b367a87893351dd599be7a27d02348947bc764721fd11cd53141
+oid sha256:7777fcad31e32edd413c6e43be6a32bc8b7a3fd8ca1f25f59c801fd09467e49d
 size 4400252568