Run 11. Outer Step 228. Inner Step 9.

config.json

@@ -8,55 +8,55 @@
     "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
+    "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "SUCCESS",
-    "107": "NON_PARTICIPATING",
+    "107": "SUCCESS",
     "108": "SUCCESS",
-    "109": "NOT_ALIVE",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
-    "110": "NON_PARTICIPATING",
+    "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
-    "116": "NON_PARTICIPATING",
+    "115": "SUCCESS",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
-    "121": "NOT_ALIVE",
+    "120": "SUCCESS",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
+    "127": "SUCCESS",
     "128": "SUCCESS",
     "129": "NOT_ALIVE",
-    "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "13": "NON_PARTICIPATING",
+    "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "134": "NON_PARTICIPATING",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "SUCCESS",
-    "141": "NOT_ALIVE",
+    "140": "NOT_ALIVE",
+    "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "SUCCESS",
     "147": "SUCCESS",
-    "148": "NON_PARTICIPATING",
+    "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
@@ -66,7 +66,7 @@
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "NOT_ALIVE",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
@@ -75,29 +75,29 @@
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
-    "165": "NOT_ALIVE",
+    "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NOT_ALIVE",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NON_PARTICIPATING",
-    "175": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "SUCCESS",
-    "179": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
-    "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
+    "184": "NOT_ALIVE",
+    "185": "NON_PARTICIPATING",
     "186": "SUCCESS",
     "187": "SUCCESS",
     "188": "NOT_ALIVE",
@@ -112,19 +112,19 @@
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "SUCCESS",
+    "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
-    "200": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
+    "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
+    "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
@@ -137,28 +137,28 @@
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "SUCCESS",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "NOT_ALIVE",
+    "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
-    "230": "SUCCESS",
+    "23": "SUCCESS",
+    "230": "NON_PARTICIPATING",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "SUCCESS",
+    "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
     "237": "SUCCESS",
-    "238": "SUCCESS",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "SUCCESS",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "SUCCESS",
     "242": "NOT_ALIVE",
@@ -166,17 +166,17 @@
     "244": "NOT_ALIVE",
     "245": "SUCCESS",
     "246": "NOT_ALIVE",
-    "247": "NOT_ALIVE",
+    "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "NOT_ALIVE",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "SUCCESS",
+    "26": "NOT_ALIVE",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -185,13 +185,13 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "SUCCESS",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
-    "38": "NON_PARTICIPATING",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
-    "4": "NON_PARTICIPATING",
+    "4": "SUCCESS",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
@@ -204,7 +204,7 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
@@ -214,19 +214,19 @@
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NOT_ALIVE",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -234,7 +234,7 @@
     "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "79": "NOT_ALIVE",
     "8": "SUCCESS",
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
@@ -253,9 +253,9 @@
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "96": "SUCCESS",
     "97": "SUCCESS",
-    "98": "SUCCESS",
+    "98": "NON_PARTICIPATING",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6669998,
-    6670027
+    6670111,
+    6670140
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6669670,
+  "last_allreduce_block": 6670047,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

gradients.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:503cd078f8905fabb029866f202183c22612e0dcf206c41c8f105f242fa318bd
+oid sha256:cad7b16d560d90a416da3f1c2f763e6530d5c86f0d9a3f6dfcc0746a6773d427
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a1d482dc7ca84eec3de8c22f834e57cb7509f7fd8bbe538976689c7a3a26e1e
+oid sha256:596c1b28ddd766b135942d9d437de9fee4fd111f17880d811cda30cc84dcf8de
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da67145f6054b030ceeb061a425f541c2f4181bcd9e965713f1f12eede3f6767
+oid sha256:3217d94769b164ab6c5240f7105b097948e29870cf3e9bdaa51b1a0823ec78ff
 size 4400216536