Run 8. Outer Step 51. Inner Step 5.

config.json

@@ -8,10 +8,10 @@
     "101": "NOT_ALIVE",
     "102": "NON_PARTICIPATING",
     "103": "NOT_ALIVE",
-    "104": "FAIL",
+    "104": "NON_PARTICIPATING",
     "105": "NOT_ALIVE",
     "106": "SUCCESS",
-    "107": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
     "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
@@ -20,7 +20,7 @@
     "112": "SUCCESS",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
@@ -32,19 +32,19 @@
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
-    "126": "FAIL",
-    "127": "FAIL",
+    "126": "SUCCESS",
+    "127": "SUCCESS",
     "128": "SUCCESS",
     "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NON_PARTICIPATING",
     "132": "SUCCESS",
-    "133": "NOT_ALIVE",
+    "133": "SUCCESS",
     "134": "NOT_ALIVE",
     "135": "SUCCESS",
     "136": "SUCCESS",
-    "137": "FAIL",
+    "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
@@ -52,7 +52,7 @@
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
-    "144": "NON_PARTICIPATING",
+    "144": "SUCCESS",
     "145": "SUCCESS",
     "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
@@ -62,7 +62,7 @@
     "150": "NOT_ALIVE",
     "151": "NON_PARTICIPATING",
     "152": "FAIL",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NON_PARTICIPATING",
     "155": "FAIL",
     "156": "NOT_ALIVE",
@@ -73,7 +73,7 @@
     "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
-    "163": "SUCCESS",
+    "163": "FAIL",
     "164": "FAIL",
     "165": "SUCCESS",
     "166": "NOT_ALIVE",
@@ -82,14 +82,14 @@
     "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "SUCCESS",
-    "171": "NON_PARTICIPATING",
-    "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
-    "174": "SUCCESS",
-    "175": "NOT_ALIVE",
+    "171": "SUCCESS",
+    "172": "NON_PARTICIPATING",
+    "173": "NON_PARTICIPATING",
+    "174": "NON_PARTICIPATING",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NON_PARTICIPATING",
-    "178": "NOT_ALIVE",
+    "178": "NON_PARTICIPATING",
     "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
@@ -98,12 +98,12 @@
     "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
-    "186": "SUCCESS",
+    "186": "FAIL",
     "187": "NOT_ALIVE",
     "188": "FAIL",
-    "189": "NON_PARTICIPATING",
+    "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "FAIL",
+    "190": "NON_PARTICIPATING",
     "191": "FAIL",
     "192": "FAIL",
     "193": "NOT_ALIVE",
@@ -115,46 +115,46 @@
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "NOT_ALIVE",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
+    "202": "FAIL",
     "203": "NON_PARTICIPATING",
     "204": "NON_PARTICIPATING",
     "205": "NOT_ALIVE",
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
+    "209": "SUCCESS",
     "21": "SUCCESS",
     "210": "NOT_ALIVE",
     "211": "NON_PARTICIPATING",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "FAIL",
-    "215": "SUCCESS",
+    "214": "NON_PARTICIPATING",
+    "215": "FAIL",
     "216": "SUCCESS",
     "217": "SUCCESS",
     "218": "FAIL",
-    "219": "FAIL",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "FAIL",
     "221": "SUCCESS",
-    "222": "NON_PARTICIPATING",
+    "222": "SUCCESS",
     "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
-    "227": "NOT_ALIVE",
+    "227": "NON_PARTICIPATING",
     "228": "SUCCESS",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
-    "234": "SUCCESS",
+    "234": "FAIL",
     "235": "NOT_ALIVE",
-    "236": "NON_PARTICIPATING",
+    "236": "SUCCESS",
     "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
@@ -163,34 +163,34 @@
     "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
-    "244": "NON_PARTICIPATING",
+    "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
-    "248": "FAIL",
+    "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
-    "252": "SUCCESS",
-    "253": "SUCCESS",
+    "251": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
+    "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
     "26": "FAIL",
     "27": "SUCCESS",
     "28": "NON_PARTICIPATING",
-    "29": "SUCCESS",
+    "29": "NOT_ALIVE",
     "3": "FAIL",
-    "30": "SUCCESS",
-    "31": "NON_PARTICIPATING",
+    "30": "NON_PARTICIPATING",
+    "31": "SUCCESS",
     "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NON_PARTICIPATING",
     "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "SUCCESS",
-    "39": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
     "41": "FAIL",
@@ -198,20 +198,20 @@
     "43": "NON_PARTICIPATING",
     "44": "FAIL",
     "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
+    "46": "SUCCESS",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
-    "5": "NOT_ALIVE",
+    "5": "FAIL",
     "50": "NOT_ALIVE",
     "51": "NON_PARTICIPATING",
-    "52": "SUCCESS",
-    "53": "SUCCESS",
-    "54": "NON_PARTICIPATING",
+    "52": "NON_PARTICIPATING",
+    "53": "FAIL",
+    "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
     "57": "NOT_ALIVE",
-    "58": "SUCCESS",
+    "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NOT_ALIVE",
@@ -234,12 +234,12 @@
     "76": "NOT_ALIVE",
     "77": "SUCCESS",
     "78": "NON_PARTICIPATING",
-    "79": "NOT_ALIVE",
+    "79": "SUCCESS",
     "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
     "81": "NON_PARTICIPATING",
-    "82": "NON_PARTICIPATING",
-    "83": "NON_PARTICIPATING",
+    "82": "SUCCESS",
+    "83": "NOT_ALIVE",
     "84": "FAIL",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -253,9 +253,9 @@
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
-    "97": "FAIL",
-    "98": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
+    "97": "NOT_ALIVE",
+    "98": "FAIL",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,18 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6374695,
-    6374712,
-    6374730
+    6374862,
+    6374877
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 78,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6374115,
+  "last_allreduce_block": 6374792,
   "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 sha256:8d83923a7263bc0cbfca2229f24905ac8c7ccf054484936b25961697d7479ed1
+oid sha256:58253f38299e95d04ece50c20c8e58578e22cc01306edd70903e3dc81103ea3b
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3484af6cfa98e3611334cb65ad586958f98bf4d8ab197fe3a8238da536810f2
+oid sha256:ad431caa3d33a5f02061fc6ce73ce403549c38ad0461251a761d173a27e567ae
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf617b19c4f89df2c3eca775596e81003ddf808ac61b93ee795212bec98530f6
+oid sha256:ea716fd3c4a1c20e13f608c2202f97650e633d4ac021a34ef155fe9db21ca0a6
 size 4400216536