Run 10. Outer Step 63. Inner Step 4.

config.json

@@ -5,257 +5,257 @@
     "1": "NON_PARTICIPATING",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
-    "102": "NOT_ALIVE",
-    "103": "SUCCESS",
+    "101": "FAIL",
+    "102": "FAIL",
+    "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "SUCCESS",
-    "106": "SUCCESS",
+    "105": "NOT_ALIVE",
+    "106": "FAIL",
     "107": "NON_PARTICIPATING",
-    "108": "NON_PARTICIPATING",
-    "109": "SUCCESS",
+    "108": "NOT_ALIVE",
+    "109": "FAIL",
     "11": "SUCCESS",
-    "110": "NON_PARTICIPATING",
-    "111": "SUCCESS",
+    "110": "NOT_ALIVE",
+    "111": "FAIL",
     "112": "NOT_ALIVE",
-    "113": "SUCCESS",
-    "114": "NOT_ALIVE",
+    "113": "NOT_ALIVE",
+    "114": "FAIL",
     "115": "NON_PARTICIPATING",
-    "116": "SUCCESS",
-    "117": "NOT_ALIVE",
-    "118": "SUCCESS",
-    "119": "NOT_ALIVE",
+    "116": "NOT_ALIVE",
+    "117": "NON_PARTICIPATING",
+    "118": "FAIL",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NON_PARTICIPATING",
-    "122": "NOT_ALIVE",
-    "123": "SUCCESS",
+    "122": "SUCCESS",
+    "123": "FAIL",
     "124": "NON_PARTICIPATING",
-    "125": "NOT_ALIVE",
-    "126": "SUCCESS",
+    "125": "FAIL",
+    "126": "FAIL",
     "127": "NON_PARTICIPATING",
-    "128": "NON_PARTICIPATING",
-    "129": "NON_PARTICIPATING",
-    "13": "SUCCESS",
+    "128": "SUCCESS",
+    "129": "SUCCESS",
+    "13": "NON_PARTICIPATING",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
-    "135": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
     "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
-    "139": "NON_PARTICIPATING",
+    "138": "NOT_ALIVE",
+    "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
-    "141": "NOT_ALIVE",
+    "141": "NON_PARTICIPATING",
     "142": "SUCCESS",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
     "146": "SUCCESS",
     "147": "SUCCESS",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
-    "15": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
+    "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
-    "154": "NOT_ALIVE",
-    "155": "SUCCESS",
+    "153": "NOT_ALIVE",
+    "154": "SUCCESS",
+    "155": "NOT_ALIVE",
     "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
-    "159": "SUCCESS",
+    "159": "FAIL",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
-    "161": "NON_PARTICIPATING",
+    "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "SUCCESS",
+    "163": "FAIL",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "SUCCESS",
+    "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "SUCCESS",
+    "169": "NON_PARTICIPATING",
     "17": "NON_PARTICIPATING",
     "170": "SUCCESS",
-    "171": "SUCCESS",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "SUCCESS",
+    "175": "FAIL",
     "176": "NOT_ALIVE",
     "177": "NON_PARTICIPATING",
-    "178": "NOT_ALIVE",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
+    "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NON_PARTICIPATING",
-    "183": "SUCCESS",
-    "184": "SUCCESS",
+    "183": "NOT_ALIVE",
+    "184": "FAIL",
     "185": "SUCCESS",
-    "186": "SUCCESS",
+    "186": "NON_PARTICIPATING",
     "187": "NON_PARTICIPATING",
-    "188": "NON_PARTICIPATING",
+    "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
-    "19": "NON_PARTICIPATING",
+    "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NOT_ALIVE",
+    "193": "NON_PARTICIPATING",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
-    "197": "NON_PARTICIPATING",
-    "198": "SUCCESS",
-    "199": "NON_PARTICIPATING",
-    "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
-    "200": "NOT_ALIVE",
+    "197": "NOT_ALIVE",
+    "198": "FAIL",
+    "199": "SUCCESS",
+    "2": "NOT_ALIVE",
+    "20": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
-    "203": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "SUCCESS",
-    "206": "NOT_ALIVE",
+    "206": "FAIL",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
-    "21": "NON_PARTICIPATING",
+    "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
-    "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
-    "213": "NON_PARTICIPATING",
-    "214": "SUCCESS",
+    "211": "NON_PARTICIPATING",
+    "212": "NON_PARTICIPATING",
+    "213": "NOT_ALIVE",
+    "214": "NOT_ALIVE",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
-    "218": "NON_PARTICIPATING",
+    "218": "NOT_ALIVE",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
-    "221": "SUCCESS",
-    "222": "SUCCESS",
-    "223": "NON_PARTICIPATING",
+    "221": "FAIL",
+    "222": "NON_PARTICIPATING",
+    "223": "FAIL",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "227": "FAIL",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "230": "FAIL",
+    "231": "NOT_ALIVE",
     "232": "SUCCESS",
     "233": "NON_PARTICIPATING",
-    "234": "SUCCESS",
-    "235": "SUCCESS",
+    "234": "NON_PARTICIPATING",
+    "235": "FAIL",
     "236": "NOT_ALIVE",
-    "237": "SUCCESS",
-    "238": "NOT_ALIVE",
+    "237": "NON_PARTICIPATING",
+    "238": "NON_PARTICIPATING",
     "239": "NON_PARTICIPATING",
-    "24": "NON_PARTICIPATING",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
-    "246": "SUCCESS",
+    "245": "NON_PARTICIPATING",
+    "246": "FAIL",
     "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
-    "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "249": "NOT_ALIVE",
+    "25": "NOT_ALIVE",
+    "250": "NOT_ALIVE",
     "251": "NOT_ALIVE",
-    "252": "NOT_ALIVE",
+    "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
     "254": "NON_PARTICIPATING",
     "255": "SUCCESS",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
-    "28": "SUCCESS",
-    "29": "SUCCESS",
-    "3": "SUCCESS",
+    "28": "NON_PARTICIPATING",
+    "29": "FAIL",
+    "3": "FAIL",
     "30": "NON_PARTICIPATING",
-    "31": "SUCCESS",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
     "35": "SUCCESS",
-    "36": "NON_PARTICIPATING",
+    "36": "NOT_ALIVE",
     "37": "NON_PARTICIPATING",
-    "38": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "SUCCESS",
     "40": "NON_PARTICIPATING",
-    "41": "NOT_ALIVE",
-    "42": "NON_PARTICIPATING",
+    "41": "NON_PARTICIPATING",
+    "42": "FAIL",
     "43": "NOT_ALIVE",
-    "44": "SUCCESS",
+    "44": "FAIL",
     "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NOT_ALIVE",
-    "49": "NON_PARTICIPATING",
-    "5": "NON_PARTICIPATING",
+    "49": "NOT_ALIVE",
+    "5": "NOT_ALIVE",
     "50": "SUCCESS",
-    "51": "NON_PARTICIPATING",
-    "52": "NON_PARTICIPATING",
+    "51": "FAIL",
+    "52": "SUCCESS",
     "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
-    "57": "SUCCESS",
+    "57": "FAIL",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "FAIL",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
-    "62": "NOT_ALIVE",
+    "60": "FAIL",
+    "61": "FAIL",
+    "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
-    "64": "SUCCESS",
-    "65": "NOT_ALIVE",
+    "64": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
-    "7": "SUCCESS",
+    "7": "FAIL",
     "70": "NON_PARTICIPATING",
-    "71": "NON_PARTICIPATING",
-    "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
-    "74": "SUCCESS",
-    "75": "NON_PARTICIPATING",
+    "71": "NOT_ALIVE",
+    "72": "NOT_ALIVE",
+    "73": "NOT_ALIVE",
+    "74": "FAIL",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
     "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "NOT_ALIVE",
+    "82": "FAIL",
     "83": "NOT_ALIVE",
     "84": "SUCCESS",
     "85": "NON_PARTICIPATING",
     "86": "SUCCESS",
-    "87": "NOT_ALIVE",
+    "87": "FAIL",
     "88": "NOT_ALIVE",
-    "89": "SUCCESS",
+    "89": "FAIL",
     "9": "NON_PARTICIPATING",
-    "90": "SUCCESS",
+    "90": "NOT_ALIVE",
     "91": "NON_PARTICIPATING",
-    "92": "SUCCESS",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
-    "96": "SUCCESS",
+    "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
+    "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6497729,
-    6497745
+    6501495,
+    6501512
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 77,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6496834,
+  "last_allreduce_block": 6500846,
   "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 sha256:0015dbfe66ba9467d2e38f1680829a6ecb8006d6bcd8dc8c33d71f3f077c64d1
+oid sha256:858f5dac457be27baaf36ee066f599c8c3157c8d01b5259c211e9d28dbe5679b
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a866372bb516f1001d9516cb5e11eab9c936d9c6945e0a4f886133bfd9b14b27
+oid sha256:9294b3c054e7803a33dffdda52c07483c3986d2ce4d9282d8837874b6b2388ce
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6aa782f5b3976b28c6078ad1d7146e06c25cd722d97e63fdff25c1c719f9c26
+oid sha256:b4eb7ed8a123f5169b5a262898e7b0795a38ef93815362917bb351beaa6c2d08
 size 4400216536