Run 10. Outer Step 21. Inner Step 6.

config.json

@@ -3,146 +3,146 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "FAIL",
-    "10": "NON_PARTICIPATING",
+    "10": "FAIL",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "FAIL",
+    "102": "SUCCESS",
     "103": "FAIL",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "FAIL",
-    "11": "NON_PARTICIPATING",
-    "110": "NOT_ALIVE",
+    "109": "NOT_ALIVE",
+    "11": "SUCCESS",
+    "110": "NON_PARTICIPATING",
     "111": "SUCCESS",
     "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
-    "114": "FAIL",
+    "113": "SUCCESS",
+    "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
-    "119": "NON_PARTICIPATING",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
-    "121": "NOT_ALIVE",
-    "122": "NOT_ALIVE",
-    "123": "FAIL",
+    "121": "NON_PARTICIPATING",
+    "122": "NON_PARTICIPATING",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
-    "125": "NOT_ALIVE",
-    "126": "FAIL",
+    "125": "NON_PARTICIPATING",
+    "126": "SUCCESS",
     "127": "NON_PARTICIPATING",
-    "128": "NOT_ALIVE",
-    "129": "NON_PARTICIPATING",
-    "13": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
+    "13": "FAIL",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
-    "134": "NOT_ALIVE",
-    "135": "NOT_ALIVE",
+    "134": "NON_PARTICIPATING",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
-    "139": "NOT_ALIVE",
+    "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "FAIL",
-    "143": "NON_PARTICIPATING",
+    "142": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
-    "146": "FAIL",
-    "147": "NOT_ALIVE",
+    "146": "NOT_ALIVE",
+    "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
-    "15": "FAIL",
-    "150": "NON_PARTICIPATING",
+    "15": "NON_PARTICIPATING",
+    "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
-    "153": "FAIL",
-    "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
-    "156": "FAIL",
+    "153": "NOT_ALIVE",
+    "154": "SUCCESS",
+    "155": "NON_PARTICIPATING",
+    "156": "NOT_ALIVE",
     "157": "NON_PARTICIPATING",
     "158": "NOT_ALIVE",
-    "159": "NOT_ALIVE",
+    "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "FAIL",
+    "163": "SUCCESS",
     "164": "NOT_ALIVE",
-    "165": "NON_PARTICIPATING",
+    "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
-    "168": "FAIL",
-    "169": "FAIL",
-    "17": "NOT_ALIVE",
+    "168": "NOT_ALIVE",
+    "169": "SUCCESS",
+    "17": "NON_PARTICIPATING",
     "170": "FAIL",
-    "171": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NON_PARTICIPATING",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "SUCCESS",
     "178": "NOT_ALIVE",
     "179": "NON_PARTICIPATING",
-    "18": "NON_PARTICIPATING",
-    "180": "NON_PARTICIPATING",
+    "18": "SUCCESS",
+    "180": "SUCCESS",
     "181": "NOT_ALIVE",
-    "182": "NOT_ALIVE",
-    "183": "FAIL",
-    "184": "FAIL",
+    "182": "NON_PARTICIPATING",
+    "183": "SUCCESS",
+    "184": "NOT_ALIVE",
     "185": "NOT_ALIVE",
     "186": "SUCCESS",
     "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
+    "188": "SUCCESS",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
-    "190": "FAIL",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NOT_ALIVE",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
-    "196": "FAIL",
+    "195": "SUCCESS",
+    "196": "SUCCESS",
     "197": "NON_PARTICIPATING",
-    "198": "FAIL",
+    "198": "SUCCESS",
     "199": "NON_PARTICIPATING",
-    "2": "FAIL",
-    "20": "FAIL",
+    "2": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
     "200": "FAIL",
     "201": "NON_PARTICIPATING",
-    "202": "NOT_ALIVE",
+    "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "FAIL",
-    "206": "FAIL",
+    "205": "NON_PARTICIPATING",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
-    "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
+    "21": "NON_PARTICIPATING",
+    "210": "NOT_ALIVE",
     "211": "NON_PARTICIPATING",
-    "212": "FAIL",
+    "212": "SUCCESS",
     "213": "SUCCESS",
-    "214": "NOT_ALIVE",
+    "214": "SUCCESS",
     "215": "NON_PARTICIPATING",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
-    "219": "FAIL",
+    "219": "SUCCESS",
     "22": "NON_PARTICIPATING",
     "220": "NOT_ALIVE",
-    "221": "FAIL",
-    "222": "FAIL",
-    "223": "NON_PARTICIPATING",
+    "221": "SUCCESS",
+    "222": "NOT_ALIVE",
+    "223": "SUCCESS",
     "224": "NOT_ALIVE",
-    "225": "NON_PARTICIPATING",
+    "225": "SUCCESS",
     "226": "NON_PARTICIPATING",
     "227": "SUCCESS",
     "228": "NOT_ALIVE",
@@ -151,70 +151,70 @@
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
-    "233": "FAIL",
-    "234": "NON_PARTICIPATING",
-    "235": "FAIL",
-    "236": "FAIL",
-    "237": "FAIL",
-    "238": "FAIL",
+    "233": "NOT_ALIVE",
+    "234": "SUCCESS",
+    "235": "NON_PARTICIPATING",
+    "236": "NON_PARTICIPATING",
+    "237": "SUCCESS",
+    "238": "NON_PARTICIPATING",
     "239": "NON_PARTICIPATING",
-    "24": "FAIL",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
     "241": "FAIL",
     "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
-    "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
+    "248": "FAIL",
+    "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "NOT_ALIVE",
     "252": "FAIL",
     "253": "NOT_ALIVE",
-    "254": "FAIL",
-    "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
-    "28": "NON_PARTICIPATING",
-    "29": "FAIL",
-    "3": "NON_PARTICIPATING",
+    "254": "NON_PARTICIPATING",
+    "255": "SUCCESS",
+    "26": "SUCCESS",
+    "27": "NOT_ALIVE",
+    "28": "SUCCESS",
+    "29": "SUCCESS",
+    "3": "FAIL",
     "30": "NON_PARTICIPATING",
     "31": "NOT_ALIVE",
-    "32": "NOT_ALIVE",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
-    "35": "FAIL",
+    "35": "NOT_ALIVE",
     "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "NON_PARTICIPATING",
-    "39": "FAIL",
+    "38": "FAIL",
+    "39": "NOT_ALIVE",
     "4": "FAIL",
     "40": "NON_PARTICIPATING",
     "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "FAIL",
-    "45": "NOT_ALIVE",
-    "46": "FAIL",
-    "47": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
+    "46": "SUCCESS",
+    "47": "SUCCESS",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
     "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
+    "51": "NOT_ALIVE",
     "52": "NON_PARTICIPATING",
-    "53": "FAIL",
+    "53": "NON_PARTICIPATING",
     "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
-    "56": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
     "57": "FAIL",
-    "58": "NOT_ALIVE",
+    "58": "NON_PARTICIPATING",
     "59": "NON_PARTICIPATING",
     "6": "NOT_ALIVE",
-    "60": "FAIL",
+    "60": "SUCCESS",
     "61": "NOT_ALIVE",
     "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
@@ -223,37 +223,37 @@
     "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "FAIL",
+    "69": "SUCCESS",
     "7": "FAIL",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
-    "72": "NON_PARTICIPATING",
+    "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
-    "74": "FAIL",
+    "74": "SUCCESS",
     "75": "NON_PARTICIPATING",
-    "76": "NON_PARTICIPATING",
+    "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
-    "78": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
     "79": "NON_PARTICIPATING",
-    "8": "NON_PARTICIPATING",
+    "8": "SUCCESS",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NOT_ALIVE",
-    "83": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
+    "83": "NON_PARTICIPATING",
     "84": "FAIL",
     "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
-    "87": "FAIL",
+    "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "SUCCESS",
-    "90": "FAIL",
-    "91": "FAIL",
-    "92": "NON_PARTICIPATING",
+    "90": "NON_PARTICIPATING",
+    "91": "NOT_ALIVE",
+    "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
+    "96": "SUCCESS",
     "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6479141,
-    6479160
+    6479209,
+    6479230
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 33,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6478819,
+  "last_allreduce_block": 6479163,
   "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 sha256:26624cc6c6fcb7fc8f1bdaeacc5100ab2e61142f4ae88b382e0933d34f5b6060
+oid sha256:3c5b6776a60870769d072a827b7df8e26735de7f0b7da471edb0b92a1839e7a5
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7f77a1bf1b7c03c07c877637b051e83142ed4b2ad89973b28a7ceaa4938b9af
+oid sha256:cf5b223bfafe909daf5345bd60ec1f82dc874ae0013bcb1faa0bf7a7be9c9a26
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14bd45bfc736f8217ff3850c6db3f2aab7e495c5e45de238cbfd8fac449d1298
+oid sha256:c0ec042b29d1882bf02ab32cfeeac24b3c1aa4fd900b48966c7e810c680c8024
 size 4400216536