Run 11. Outer Step 383. Inner Step 9.

config.json

@@ -11,38 +11,38 @@
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NOT_ALIVE",
-    "108": "SUCCESS",
-    "109": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "112": "NON_PARTICIPATING",
+    "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
+    "115": "FAIL",
     "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
+    "128": "SUCCESS",
     "129": "NOT_ALIVE",
-    "13": "SUCCESS",
-    "130": "NON_PARTICIPATING",
+    "13": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "SUCCESS",
+    "134": "SUCCESS",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -51,18 +51,18 @@
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "SUCCESS",
+    "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -72,7 +72,7 @@
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "SUCCESS",
+    "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
@@ -94,11 +94,11 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "SUCCESS",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
-    "186": "SUCCESS",
+    "186": "NON_PARTICIPATING",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -108,10 +108,10 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
+    "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
+    "198": "SUCCESS",
     "199": "SUCCESS",
     "2": "NON_PARTICIPATING",
     "20": "SUCCESS",
@@ -133,7 +133,7 @@
     "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "217": "NOT_ALIVE",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
@@ -149,13 +149,13 @@
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "SUCCESS",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "NON_PARTICIPATING",
+    "237": "SUCCESS",
     "238": "NON_PARTICIPATING",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
@@ -166,17 +166,17 @@
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "NOT_ALIVE",
-    "247": "NON_PARTICIPATING",
+    "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
+    "251": "SUCCESS",
     "252": "NON_PARTICIPATING",
     "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": "NON_PARTICIPATING",
+    "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -193,14 +193,14 @@
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "NON_PARTICIPATING",
+    "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -223,10 +223,10 @@
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
-    "69": "NOT_ALIVE",
+    "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NOT_ALIVE",
+    "71": "SUCCESS",
     "72": "SUCCESS",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -236,7 +236,7 @@
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
     "8": "NON_PARTICIPATING",
-    "80": "NOT_ALIVE",
+    "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -255,7 +255,7 @@
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "SUCCESS",
-    "98": "NOT_ALIVE",
+    "98": "SUCCESS",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6735652,
-    6735682
+    6735811,
+    6735840
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 36,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6735409,
+  "last_allreduce_block": 6735770,
   "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 sha256:9669358b936c298910c08ef4020447ccd72482bfdeab8217827f1d7faf699f24
+oid sha256:d40c2ecc9ed263bfb6139a677c7cc2af10d2338ee7c7b2045a5fe2820ae15185
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b59258ddb4e9cff7b4658157caf57c06adcccffda6d0099eceda9a63f815c53e
+oid sha256:e3f7f2ac33cd9e368c711749e90ba78de087b58b8272a430db8967c11d938206
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fee8e4efe98d1dbcb071a8f53c07d511b62e6f2fd6d80a4682a25cbac77ab10
+oid sha256:de26da25e2e17d803ef0eff564f5f3b6ac27b1538841d8f02dc10c63c2ad560c
 size 4400216536