Run 11. Outer Step 408. Inner Step 9.

config.json

@@ -8,41 +8,41 @@
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
+    "104": "SUCCESS",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
+    "107": "NON_PARTICIPATING",
     "108": "SUCCESS",
-    "109": "NON_PARTICIPATING",
+    "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
-    "113": "SUCCESS",
+    "113": "NOT_ALIVE",
     "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": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NOT_ALIVE",
-    "123": "SUCCESS",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
+    "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
-    "13": "SUCCESS",
-    "130": "SUCCESS",
+    "13": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -62,17 +62,17 @@
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "NOT_ALIVE",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
@@ -84,7 +84,7 @@
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NOT_ALIVE",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
@@ -98,7 +98,7 @@
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
-    "186": "SUCCESS",
+    "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -111,7 +111,7 @@
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "FAIL",
+    "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "SUCCESS",
     "20": "SUCCESS",
@@ -124,15 +124,15 @@
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
-    "212": "NON_PARTICIPATING",
-    "213": "NOT_ALIVE",
+    "212": "SUCCESS",
+    "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
-    "215": "SUCCESS",
-    "216": "NOT_ALIVE",
+    "215": "NON_PARTICIPATING",
+    "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "SUCCESS",
@@ -148,8 +148,8 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
@@ -160,18 +160,18 @@
     "239": "NOT_ALIVE",
     "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "NOT_ALIVE",
-    "247": "NON_PARTICIPATING",
+    "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
+    "251": "SUCCESS",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -184,15 +184,15 @@
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
-    "33": "NOT_ALIVE",
+    "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
-    "37": "NOT_ALIVE",
+    "37": "SUCCESS",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "NOT_ALIVE",
+    "40": "NON_PARTICIPATING",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
@@ -207,12 +207,12 @@
     "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "NOT_ALIVE",
+    "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "NOT_ALIVE",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "NOT_ALIVE",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
@@ -224,19 +224,19 @@
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
-    "7": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
-    "71": "FAIL",
-    "72": "SUCCESS",
+    "71": "SUCCESS",
+    "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NOT_ALIVE",
+    "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "SUCCESS",
+    "79": "NOT_ALIVE",
     "8": "SUCCESS",
-    "80": "SUCCESS",
+    "80": "FAIL",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -254,7 +254,7 @@
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NON_PARTICIPATING",
+    "97": "SUCCESS",
     "98": "SUCCESS",
     "99": "SUCCESS"
   },
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6750050,
-    6750086
+    6750278,
+    6750323
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 27,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6749857,
+  "last_allreduce_block": 6750233,
   "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 sha256:57464412a000122eee2cf6d1f6570b449fc3dcddae72b8fd5a83576b8b0bcc77
+oid sha256:1073f6e7edab8e9940e237f6aa424b03e68f36ab875577ed136a25c35f902d49
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a06f6bb95329dd19248dbe9305c11746a4ddc4d1233a02f1eb08d5cf753118d
+oid sha256:27888fbdebcb9ee1c0be9b1e4882c53edd84bc701c1bb1b52ec18fc3a4329846
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87a3f569ba868ea3ff164a31333b4b3060aeb9fb9aecbcf120b3c4eed3a6dd71
+oid sha256:571fb25e6bfe7a9eab6e6976c55c84452c767b2d92d2fa8842bca5c861c92a97
 size 4400216536