Run 11. Outer Step 55. Inner Step 9.

config.json

@@ -8,7 +8,7 @@
     "101": "NOT_ALIVE",
     "102": "SUCCESS",
     "103": "NOT_ALIVE",
-    "104": "FAIL",
+    "104": "NON_PARTICIPATING",
     "105": "SUCCESS",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
@@ -20,7 +20,7 @@
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
+    "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
     "117": "SUCCESS",
     "118": "SUCCESS",
@@ -31,13 +31,13 @@
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "SUCCESS",
-    "125": "NON_PARTICIPATING",
-    "126": "NON_PARTICIPATING",
+    "125": "SUCCESS",
+    "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
     "13": "NON_PARTICIPATING",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
@@ -47,7 +47,7 @@
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "SUCCESS",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "SUCCESS",
@@ -57,7 +57,7 @@
     "146": "NOT_ALIVE",
     "147": "NOT_ALIVE",
     "148": "NOT_ALIVE",
-    "149": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
@@ -66,44 +66,44 @@
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "SUCCESS",
     "158": "SUCCESS",
-    "159": "FAIL",
-    "16": "NOT_ALIVE",
+    "159": "SUCCESS",
+    "16": "NON_PARTICIPATING",
     "160": "NOT_ALIVE",
-    "161": "NON_PARTICIPATING",
+    "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NON_PARTICIPATING",
     "167": "NOT_ALIVE",
-    "168": "SUCCESS",
+    "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "FAIL",
+    "173": "NON_PARTICIPATING",
     "174": "NON_PARTICIPATING",
-    "175": "SUCCESS",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
-    "18": "SUCCESS",
+    "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
-    "185": "SUCCESS",
+    "185": "NOT_ALIVE",
     "186": "SUCCESS",
     "187": "NOT_ALIVE",
-    "188": "NON_PARTICIPATING",
+    "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
-    "190": "NON_PARTICIPATING",
+    "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
@@ -111,22 +111,22 @@
     "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
     "197": "SUCCESS",
-    "198": "NOT_ALIVE",
+    "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NON_PARTICIPATING",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
-    "203": "NOT_ALIVE",
+    "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "NOT_ALIVE",
-    "21": "NON_PARTICIPATING",
-    "210": "NOT_ALIVE",
+    "21": "SUCCESS",
+    "210": "NON_PARTICIPATING",
     "211": "NOT_ALIVE",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
@@ -140,25 +140,25 @@
     "220": "NON_PARTICIPATING",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "SUCCESS",
+    "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "NON_PARTICIPATING",
+    "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
-    "232": "NON_PARTICIPATING",
+    "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "SUCCESS",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
-    "237": "NOT_ALIVE",
-    "238": "NON_PARTICIPATING",
+    "237": "NON_PARTICIPATING",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
     "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
@@ -167,7 +167,7 @@
     "245": "NOT_ALIVE",
     "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
-    "248": "FAIL",
+    "248": "NON_PARTICIPATING",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "SUCCESS",
@@ -176,7 +176,7 @@
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "FAIL",
+    "26": "NON_PARTICIPATING",
     "27": "SUCCESS",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -189,8 +189,8 @@
     "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
-    "38": "SUCCESS",
-    "39": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
+    "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
     "41": "SUCCESS",
@@ -205,17 +205,17 @@
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NON_PARTICIPATING",
-    "52": "NON_PARTICIPATING",
+    "52": "SUCCESS",
     "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
-    "57": "FAIL",
-    "58": "NOT_ALIVE",
+    "57": "NOT_ALIVE",
+    "58": "NON_PARTICIPATING",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "61": "SUCCESS",
     "62": "SUCCESS",
     "63": "NON_PARTICIPATING",
     "64": "SUCCESS",
@@ -223,40 +223,40 @@
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "NON_PARTICIPATING",
-    "7": "NOT_ALIVE",
+    "69": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
-    "74": "NON_PARTICIPATING",
+    "74": "NOT_ALIVE",
     "75": "SUCCESS",
     "76": "SUCCESS",
     "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "78": "NON_PARTICIPATING",
+    "79": "SUCCESS",
     "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "FAIL",
+    "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
+    "85": "SUCCESS",
     "86": "SUCCESS",
-    "87": "NOT_ALIVE",
-    "88": "NOT_ALIVE",
+    "87": "SUCCESS",
+    "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "SUCCESS",
     "91": "NOT_ALIVE",
-    "92": "SUCCESS",
-    "93": "NOT_ALIVE",
+    "92": "NOT_ALIVE",
+    "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "SUCCESS",
-    "99": "NOT_ALIVE"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6573590,
-    6573625
+    6573713,
+    6573742
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 53,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6573223,
+  "last_allreduce_block": 6573649,
   "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:c023415ccb3eac9eddabe72a582c62b25489c94a40918a5c3010d52ed0bec304
+oid sha256:7f00bc7af0994f07beed24613a7c2bd9d61ff591e885a8c52b85219df8625bd6
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5a246a941542e631e590c40fbffb3366ae38930b10f606ce499f54cd5bbfbda
+oid sha256:e34149e3c7a3e7698b950c2afbb6187b591ff845bd00ef51257301f4fc896e2d
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd4506da5568ccff5a422a9b05f60a70a62724942f6d60bfa7d65caeaa85c47e
+oid sha256:c95b5c72d3901d2ab87cb1f304debf097d3e26378cd8707d6919fa763a39d1e1
 size 4400216536