Run 11. Outer Step 181. Inner Step 9.

config.json

@@ -18,24 +18,24 @@
     "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NON_PARTICIPATING",
-    "119": "SUCCESS",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NON_PARTICIPATING",
-    "123": "SUCCESS",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
+    "127": "SUCCESS",
     "128": "NOT_ALIVE",
-    "129": "SUCCESS",
+    "129": "NON_PARTICIPATING",
     "13": "NON_PARTICIPATING",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
@@ -46,31 +46,31 @@
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
-    "14": "NOT_ALIVE",
+    "139": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
-    "142": "NON_PARTICIPATING",
+    "142": "SUCCESS",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "SUCCESS",
-    "148": "SUCCESS",
+    "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "SUCCESS",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "NOT_ALIVE",
     "158": "NOT_ALIVE",
     "159": "NON_PARTICIPATING",
-    "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "16": "NON_PARTICIPATING",
+    "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
@@ -81,7 +81,7 @@
     "168": "NOT_ALIVE",
     "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
-    "170": "NOT_ALIVE",
+    "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
@@ -89,7 +89,7 @@
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
-    "178": "NOT_ALIVE",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -97,9 +97,9 @@
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
-    "185": "SUCCESS",
+    "185": "NOT_ALIVE",
     "186": "SUCCESS",
-    "187": "NOT_ALIVE",
+    "187": "SUCCESS",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -110,9 +110,9 @@
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "NOT_ALIVE",
+    "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "SUCCESS",
+    "199": "NON_PARTICIPATING",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
     "200": "NOT_ALIVE",
@@ -128,7 +128,7 @@
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
-    "212": "NON_PARTICIPATING",
+    "212": "SUCCESS",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
@@ -140,7 +140,7 @@
     "220": "NOT_ALIVE",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "SUCCESS",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
@@ -152,44 +152,44 @@
     "231": "SUCCESS",
     "232": "NON_PARTICIPATING",
     "233": "NOT_ALIVE",
-    "234": "NON_PARTICIPATING",
+    "234": "NOT_ALIVE",
     "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
+    "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "SUCCESS",
-    "241": "SUCCESS",
+    "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
+    "245": "NOT_ALIVE",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
-    "252": "NON_PARTICIPATING",
+    "251": "SUCCESS",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "NOT_ALIVE",
     "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "NON_PARTICIPATING",
+    "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "SUCCESS",
+    "34": "NON_PARTICIPATING",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
-    "38": "NON_PARTICIPATING",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
@@ -213,20 +213,20 @@
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
-    "6": "NON_PARTICIPATING",
-    "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "6": "NOT_ALIVE",
+    "60": "NOT_ALIVE",
+    "61": "NOT_ALIVE",
     "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "63": "SUCCESS",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "NON_PARTICIPATING",
+    "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
+    "71": "FAIL",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -235,8 +235,8 @@
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
-    "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "8": "NOT_ALIVE",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -251,11 +251,11 @@
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "SUCCESS",
     "97": "NOT_ALIVE",
-    "98": "SUCCESS",
+    "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6650560,
-    6650590
+    6650758,
+    6650789
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 50,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6649919,
+  "last_allreduce_block": 6650656,
   "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 sha256:4e585344ce8bf1b47249b0506b8aa2068aa7b23caf4a072785fedbb972df1655
+oid sha256:384277c97ce32784189fbb5df6db19576da0f10e3433ed2745c9442945fcc032
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:167e707cc53a8dd42522ac3138d88e30402c54dce9bf41f63b44bfe88e55b191
+oid sha256:5f38fdd6477fdd6f1192036823b6bf5b30afa50f97086325331ab2ff5bb1a312
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac97f49aa39c2e571d62a6e8c8567939aad1be361d924c3e064b9cc32c4787ab
+oid sha256:3f3a7f94331ba83fd2e37af0da805bf754a5c01dc774156b522945ef782d28b4
 size 4400216536