Run 11. Outer Step 58. Inner Step 9.

config.json

@@ -8,9 +8,9 @@
     "101": "NOT_ALIVE",
     "102": "SUCCESS",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
+    "104": "SUCCESS",
     "105": "SUCCESS",
-    "106": "SUCCESS",
+    "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
     "109": "SUCCESS",
@@ -29,9 +29,9 @@
     "120": "NON_PARTICIPATING",
     "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "SUCCESS",
-    "125": "NON_PARTICIPATING",
+    "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
@@ -40,24 +40,24 @@
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "FAIL",
+    "133": "NON_PARTICIPATING",
+    "134": "NON_PARTICIPATING",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
-    "137": "NOT_ALIVE",
+    "137": "NON_PARTICIPATING",
     "138": "NOT_ALIVE",
     "139": "SUCCESS",
     "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "SUCCESS",
+    "142": "NON_PARTICIPATING",
     "143": "SUCCESS",
     "144": "NOT_ALIVE",
-    "145": "NOT_ALIVE",
+    "145": "NON_PARTICIPATING",
     "146": "NOT_ALIVE",
     "147": "NOT_ALIVE",
     "148": "NOT_ALIVE",
-    "149": "NON_PARTICIPATING",
+    "149": "SUCCESS",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
@@ -66,17 +66,17 @@
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
-    "158": "NOT_ALIVE",
-    "159": "FAIL",
-    "16": "NON_PARTICIPATING",
-    "160": "NOT_ALIVE",
+    "157": "NON_PARTICIPATING",
+    "158": "SUCCESS",
+    "159": "NOT_ALIVE",
+    "16": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
-    "162": "NOT_ALIVE",
-    "163": "NON_PARTICIPATING",
+    "162": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
+    "166": "SUCCESS",
     "167": "NOT_ALIVE",
     "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
@@ -97,28 +97,28 @@
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
+    "185": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
-    "19": "NON_PARTICIPATING",
+    "19": "SUCCESS",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
-    "196": "NOT_ALIVE",
+    "196": "NON_PARTICIPATING",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
-    "20": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
-    "203": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NON_PARTICIPATING",
     "206": "SUCCESS",
@@ -126,11 +126,11 @@
     "208": "NOT_ALIVE",
     "209": "NOT_ALIVE",
     "21": "NON_PARTICIPATING",
-    "210": "NON_PARTICIPATING",
+    "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
-    "214": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
     "217": "SUCCESS",
@@ -152,7 +152,7 @@
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
-    "234": "SUCCESS",
+    "234": "NOT_ALIVE",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "SUCCESS",
@@ -172,7 +172,7 @@
     "25": "SUCCESS",
     "250": "SUCCESS",
     "251": "NOT_ALIVE",
-    "252": "NOT_ALIVE",
+    "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
@@ -180,25 +180,25 @@
     "27": "SUCCESS",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
-    "3": "SUCCESS",
+    "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "SUCCESS",
-    "35": "NON_PARTICIPATING",
+    "35": "SUCCESS",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
     "38": "SUCCESS",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "SUCCESS",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
-    "46": "SUCCESS",
+    "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "SUCCESS",
     "49": "SUCCESS",
@@ -207,29 +207,29 @@
     "51": "NON_PARTICIPATING",
     "52": "SUCCESS",
     "53": "SUCCESS",
-    "54": "NOT_ALIVE",
+    "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
-    "57": "NOT_ALIVE",
-    "58": "NON_PARTICIPATING",
+    "57": "SUCCESS",
+    "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
+    "60": "NOT_ALIVE",
+    "61": "NOT_ALIVE",
     "62": "SUCCESS",
     "63": "NOT_ALIVE",
-    "64": "SUCCESS",
+    "64": "NOT_ALIVE",
     "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
-    "7": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
-    "74": "NOT_ALIVE",
+    "74": "NON_PARTICIPATING",
     "75": "SUCCESS",
     "76": "SUCCESS",
     "77": "NOT_ALIVE",
@@ -239,21 +239,21 @@
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "NOT_ALIVE",
+    "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
-    "86": "NOT_ALIVE",
+    "85": "NOT_ALIVE",
+    "86": "NON_PARTICIPATING",
     "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "SUCCESS",
-    "91": "NOT_ALIVE",
+    "91": "NON_PARTICIPATING",
     "92": "SUCCESS",
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "SUCCESS",
     "99": "SUCCESS"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6574915,
-    6574950
+    6575014,
+    6575043
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 58,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6574492,
+  "last_allreduce_block": 6574967,
   "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:b24638f6559fb3443fc57469e04539b46b10563209b6094fdec9b8941b58814a
+oid sha256:085cc375a8817cdc690daed58af90cffa9f3836448950438e502939f2469d4b8
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d82b37ea23cdc38ea566da7ee585e0964dc554b3a4dd4adffa18689652982a8b
+oid sha256:62314ff1ca7b24cb636cab8671a6bfa93b641ec12b2378d9372bafd24622c7d2
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71a5ca4003861e2567c777280ed434aa5f57f1c42a06d97f3b357bcb1dc396d6
+oid sha256:0f20af6f23f0159c4ace9b2f78392057e75f2f0dff26d7136d275d7f2a321022
 size 4400216536