Run 7. Outer Step 54. Inner Step 0. Peers 30.

config.json

@@ -5,7 +5,7 @@
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
     "100": "NON_PARTICIPATING",
-    "101": "FAIL",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "NOT_ALIVE",
@@ -17,19 +17,19 @@
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "NOT_ALIVE",
-    "112": "NON_PARTICIPATING",
+    "112": "SUCCESS",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NON_PARTICIPATING",
-    "119": "NOT_ALIVE",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
@@ -37,29 +37,29 @@
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
-    "130": "SUCCESS",
+    "130": "FAIL",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "SUCCESS",
     "134": "NOT_ALIVE",
-    "135": "SUCCESS",
-    "136": "FAIL",
-    "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
+    "136": "NON_PARTICIPATING",
+    "137": "NON_PARTICIPATING",
+    "138": "SUCCESS",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
-    "143": "SUCCESS",
+    "143": "NON_PARTICIPATING",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "SUCCESS",
+    "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
@@ -70,18 +70,18 @@
     "158": "SUCCESS",
     "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
-    "160": "NON_PARTICIPATING",
+    "160": "NOT_ALIVE",
     "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
     "163": "NON_PARTICIPATING",
-    "164": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
-    "166": "SUCCESS",
+    "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
     "169": "SUCCESS",
     "17": "NOT_ALIVE",
-    "170": "NON_PARTICIPATING",
+    "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
@@ -97,18 +97,18 @@
     "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
-    "185": "NOT_ALIVE",
+    "185": "NON_PARTICIPATING",
     "186": "SUCCESS",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
-    "191": "NON_PARTICIPATING",
+    "191": "NOT_ALIVE",
     "192": "SUCCESS",
     "193": "NON_PARTICIPATING",
-    "194": "FAIL",
-    "195": "NOT_ALIVE",
+    "194": "NON_PARTICIPATING",
+    "195": "SUCCESS",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
@@ -127,7 +127,7 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NON_PARTICIPATING",
-    "211": "NON_PARTICIPATING",
+    "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
@@ -135,7 +135,7 @@
     "216": "NOT_ALIVE",
     "217": "SUCCESS",
     "218": "NON_PARTICIPATING",
-    "219": "FAIL",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "SUCCESS",
@@ -145,7 +145,7 @@
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
-    "228": "SUCCESS",
+    "228": "FAIL",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
@@ -155,7 +155,7 @@
     "234": "NON_PARTICIPATING",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
-    "237": "NOT_ALIVE",
+    "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
@@ -165,7 +165,7 @@
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
-    "246": "SUCCESS",
+    "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
@@ -173,12 +173,12 @@
     "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
-    "253": "NON_PARTICIPATING",
+    "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "NOT_ALIVE",
-    "28": "NON_PARTICIPATING",
+    "27": "NON_PARTICIPATING",
+    "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
@@ -194,11 +194,11 @@
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
     "41": "NOT_ALIVE",
-    "42": "NOT_ALIVE",
+    "42": "NON_PARTICIPATING",
     "43": "NON_PARTICIPATING",
     "44": "SUCCESS",
     "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
+    "46": "SUCCESS",
     "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
@@ -206,28 +206,28 @@
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
     "52": "NOT_ALIVE",
-    "53": "NON_PARTICIPATING",
+    "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NOT_ALIVE",
     "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
-    "70": "NON_PARTICIPATING",
-    "71": "NOT_ALIVE",
-    "72": "NON_PARTICIPATING",
+    "70": "NOT_ALIVE",
+    "71": "NON_PARTICIPATING",
+    "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
@@ -240,10 +240,10 @@
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
-    "87": "SUCCESS",
+    "87": "FAIL",
     "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
@@ -255,7 +255,7 @@
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
-    "98": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6255623,
+  "last_allreduce_block": 6256319,
   "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:d37a909f9593309405ea4818c4f94c4cebf10085a3eca78d36a193225db6e2fe
+oid sha256:2fd6ee42b4ed3ee59204675726ab0d964ae62ae4474be01bae345eddbb10ebe7
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea8fb701050025df0408cb49493c0ed664422cf3a9dc5db51196280faf53ebe5
+oid sha256:007b0e246f8cc2a4eb7dee6ae82162a7593299919eb1c9d2d2fdeb399afad6ae
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c87d95d282ef21c84b4d43186d3751acdf15e66f121104a379d6f6c3dea6b2d
+oid sha256:f3cea72ef1c0fcbad14982a69555d96a56de0aa4473386e44178380d1ced08a5
 size 4400252568