Run 11. Outer Step 399. Inner Step 9.

config.json

@@ -5,10 +5,10 @@
     "1": "NOT_ALIVE",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
+    "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
+    "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "SUCCESS",
@@ -20,20 +20,20 @@
     "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
-    "116": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NOT_ALIVE",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "SUCCESS",
+    "127": "NOT_ALIVE",
     "128": "SUCCESS",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
@@ -41,14 +41,14 @@
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
+    "134": "SUCCESS",
     "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "NON_PARTICIPATING",
+    "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
@@ -62,7 +62,7 @@
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -70,7 +70,7 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NON_PARTICIPATING",
+    "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
@@ -79,7 +79,7 @@
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
@@ -94,12 +94,12 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "SUCCESS",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
+    "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -127,13 +127,13 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "SUCCESS",
-    "212": "NON_PARTICIPATING",
+    "211": "NOT_ALIVE",
+    "212": "SUCCESS",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
-    "217": "SUCCESS",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
@@ -143,7 +143,7 @@
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "NON_PARTICIPATING",
+    "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
@@ -155,7 +155,7 @@
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "NON_PARTICIPATING",
+    "237": "NOT_ALIVE",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
@@ -171,7 +171,7 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
+    "251": "NON_PARTICIPATING",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -188,7 +188,7 @@
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "NON_PARTICIPATING",
+    "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
@@ -204,17 +204,17 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NON_PARTICIPATING",
+    "51": "NOT_ALIVE",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "NON_PARTICIPATING",
+    "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
     "57": "SUCCESS",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
@@ -226,11 +226,11 @@
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
+    "71": "NOT_ALIVE",
     "72": "SUCCESS",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "SUCCESS",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
@@ -239,24 +239,24 @@
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "NOT_ALIVE",
+    "83": "SUCCESS",
     "84": "SUCCESS",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
+    "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
-    "92": "NOT_ALIVE",
+    "92": "SUCCESS",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "NOT_ALIVE"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6742025,
-    6742062
+    6742276,
+    6742303
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 31,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6741808,
+  "last_allreduce_block": 6742237,
   "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:fc6f9b33115cf02de39f4791fcdbd8f5317c06458caf1c1979ba63bb5a54f8d7
+oid sha256:3b9b2ffbfd02c8018e54040fee80d9b8d600d3909c52cd537b7397333fe09bc3
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6486fc3500afc95e547d1cdbcbac85c07cdbc263b6c9630c06182378cd8395d
+oid sha256:f6b8865ccae039d61edd0c999ea45104b944d1cdde1a41065e9b9a3c1acfc536
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d21947ce267f1449423915de6d51ec949d35a1e54338ab89d531fa62756e2e36
+oid sha256:60c5418246f28a95b21b6dfd179e912797345ca784f75128342af03b5ce55c14
 size 4400216536