Run 11. Outer Step 398. Inner Step 9.

config.json

@@ -5,26 +5,26 @@
     "1": "NOT_ALIVE",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "SUCCESS",
+    "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
+    "107": "SUCCESS",
     "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NON_PARTICIPATING",
+    "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
-    "116": "SUCCESS",
+    "115": "NOT_ALIVE",
+    "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "SUCCESS",
@@ -33,22 +33,22 @@
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
-    "128": "NON_PARTICIPATING",
+    "127": "SUCCESS",
+    "128": "SUCCESS",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
-    "130": "NON_PARTICIPATING",
+    "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
+    "134": "NOT_ALIVE",
     "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "SUCCESS",
+    "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
@@ -62,7 +62,7 @@
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -70,14 +70,14 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "SUCCESS",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
@@ -90,16 +90,16 @@
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
+    "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "NON_PARTICIPATING",
-    "186": "SUCCESS",
-    "187": "SUCCESS",
+    "185": "SUCCESS",
+    "186": "NOT_ALIVE",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -111,29 +111,29 @@
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NOT_ALIVE",
-    "199": "NOT_ALIVE",
+    "198": "NON_PARTICIPATING",
+    "199": "SUCCESS",
     "2": "SUCCESS",
     "20": "SUCCESS",
-    "200": "NON_PARTICIPATING",
+    "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
-    "212": "NOT_ALIVE",
+    "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
@@ -148,15 +148,15 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NON_PARTICIPATING",
+    "230": "SUCCESS",
     "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "SUCCESS",
-    "238": "NON_PARTICIPATING",
+    "237": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
@@ -171,12 +171,12 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "SUCCESS",
-    "252": "NON_PARTICIPATING",
+    "251": "NOT_ALIVE",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -184,7 +184,7 @@
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
-    "33": "NOT_ALIVE",
+    "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -214,19 +214,19 @@
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
     "64": "SUCCESS",
-    "65": "SUCCESS",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "FAIL",
+    "71": "SUCCESS",
     "72": "SUCCESS",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -235,17 +235,17 @@
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
-    "8": "NOT_ALIVE",
+    "8": "NON_PARTICIPATING",
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "NOT_ALIVE",
     "84": "SUCCESS",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NOT_ALIVE",
-    "89": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
@@ -254,9 +254,9 @@
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NON_PARTICIPATING",
-    "98": "NOT_ALIVE",
-    "99": "SUCCESS"
+    "97": "SUCCESS",
+    "98": "SUCCESS",
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6741645,
-    6741680
+    6741854,
+    6741883
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6741322,
+  "last_allreduce_block": 6741808,
   "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 sha256:75b7956ff9bf1eef6b9cf9a239c4632d228f350bc9a3e85d83f0c883ba0b61db
+oid sha256:bdfa98560294d79c571dca0cda93e0c93ca4de4a55fe46b858568bbc79927928
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc7ed3399cebcdbbed8a101be927f139dbdab380c2d6cdbf4e18312e2d2396e4
+oid sha256:b5a03d0165f444316273f6f0a1d245a5b4f3ab14c99c1c61907e212b3ed40881
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5236186d1cfea7aeb96b4c43c6dede0342c61a105be12d960c666347b82d4109
+oid sha256:17748a255ac83e0275bea195adddf40850724a487e311a4113acfeaad07aed08
 size 4400216536