Run 11. Outer Step 215. Inner Step 9.

config.json

@@ -3,45 +3,45 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "SUCCESS",
-    "10": "SUCCESS",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
-    "103": "NON_PARTICIPATING",
+    "103": "SUCCESS",
     "104": "NON_PARTICIPATING",
-    "105": "FAIL",
-    "106": "FAIL",
-    "107": "NOT_ALIVE",
-    "108": "FAIL",
-    "109": "FAIL",
+    "105": "NON_PARTICIPATING",
+    "106": "SUCCESS",
+    "107": "NON_PARTICIPATING",
+    "108": "SUCCESS",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
-    "116": "NOT_ALIVE",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "FAIL",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "120": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NON_PARTICIPATING",
-    "128": "NON_PARTICIPATING",
+    "127": "SUCCESS",
+    "128": "SUCCESS",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NON_PARTICIPATING",
+    "134": "SUCCESS",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
@@ -49,20 +49,20 @@
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
-    "141": "SUCCESS",
+    "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "FAIL",
+    "143": "SUCCESS",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
-    "147": "FAIL",
+    "146": "SUCCESS",
+    "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "SUCCESS",
+    "150": "FAIL",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "FAIL",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -72,7 +72,7 @@
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
@@ -84,12 +84,12 @@
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "FAIL",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "SUCCESS",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -97,26 +97,26 @@
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
-    "187": "SUCCESS",
+    "185": "NOT_ALIVE",
+    "186": "SUCCESS",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "FAIL",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "FAIL",
+    "195": "SUCCESS",
     "196": "NOT_ALIVE",
-    "197": "FAIL",
+    "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "NON_PARTICIPATING",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
     "200": "NON_PARTICIPATING",
-    "201": "NON_PARTICIPATING",
+    "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
@@ -126,10 +126,10 @@
     "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "NOT_ALIVE",
+    "210": "NON_PARTICIPATING",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
-    "213": "NOT_ALIVE",
+    "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
@@ -137,41 +137,41 @@
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "NOT_ALIVE",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "FAIL",
-    "227": "SUCCESS",
+    "226": "SUCCESS",
+    "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
     "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
-    "234": "NOT_ALIVE",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
-    "237": "NOT_ALIVE",
+    "236": "NOT_ALIVE",
+    "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
-    "246": "NON_PARTICIPATING",
+    "245": "SUCCESS",
+    "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
+    "251": "SUCCESS",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -185,16 +185,16 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
-    "38": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
-    "41": "FAIL",
-    "42": "NON_PARTICIPATING",
+    "41": "SUCCESS",
+    "42": "SUCCESS",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NON_PARTICIPATING",
@@ -204,10 +204,10 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "SUCCESS",
+    "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
@@ -222,11 +222,11 @@
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "FAIL",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NOT_ALIVE",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -236,10 +236,10 @@
     "78": "NOT_ALIVE",
     "79": "NON_PARTICIPATING",
     "8": "SUCCESS",
-    "80": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
-    "83": "FAIL",
+    "82": "NOT_ALIVE",
+    "83": "NOT_ALIVE",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -254,7 +254,7 @@
     "94": "NOT_ALIVE",
     "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": [
-    6664827,
-    6664854
+    6664962,
+    6664980
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 63,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6664526,
+  "last_allreduce_block": 6664911,
   "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 sha256:3859709ef275d58e15d130d2bba3cdd70627d5c2ff356230c46df2a192ebcdf8
+oid sha256:ad8a528e8a5fc933ac8bbd4873c740e73e9be4fe4aa09c4ce7e9e20310be5a76
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6412a36430e55d5075172bcbfdd7226fb12f046330814fd595cbd01c6c716224
+oid sha256:a83d16dd8df3582d4cee0ba69e9c3961caa6ea7a4ab5a6cf98f531c3b0531ba7
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7137485c449596f2d7251930c3f4c33916c8fb7da2587040e6ef8cfb77f33721
+oid sha256:52e33628a0d075fdd17c6a5912efb94bfe388357a0ca36f47d61369aa170f68e
 size 4400216536