Run 8. Outer Step 97. Inner Step 5.

config.json

@@ -7,255 +7,255 @@
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
-    "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
+    "103": "NON_PARTICIPATING",
+    "104": "NON_PARTICIPATING",
+    "105": "NON_PARTICIPATING",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
-    "109": "SUCCESS",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
-    "111": "NOT_ALIVE",
+    "110": "NON_PARTICIPATING",
+    "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
-    "114": "NON_PARTICIPATING",
+    "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
-    "116": "SUCCESS",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
-    "118": "NON_PARTICIPATING",
-    "119": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "SUCCESS",
-    "121": "NOT_ALIVE",
-    "122": "NON_PARTICIPATING",
-    "123": "SUCCESS",
-    "124": "SUCCESS",
-    "125": "NOT_ALIVE",
-    "126": "SUCCESS",
+    "121": "NON_PARTICIPATING",
+    "122": "NOT_ALIVE",
+    "123": "NON_PARTICIPATING",
+    "124": "NON_PARTICIPATING",
+    "125": "NON_PARTICIPATING",
+    "126": "FAIL",
     "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
-    "129": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
-    "130": "SUCCESS",
-    "131": "NON_PARTICIPATING",
+    "130": "NON_PARTICIPATING",
+    "131": "NOT_ALIVE",
     "132": "SUCCESS",
-    "133": "NOT_ALIVE",
-    "134": "SUCCESS",
+    "133": "SUCCESS",
+    "134": "NON_PARTICIPATING",
     "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
-    "137": "SUCCESS",
+    "137": "NON_PARTICIPATING",
     "138": "NOT_ALIVE",
     "139": "NOT_ALIVE",
-    "14": "NOT_ALIVE",
+    "14": "SUCCESS",
     "140": "NOT_ALIVE",
-    "141": "NON_PARTICIPATING",
-    "142": "FAIL",
+    "141": "NOT_ALIVE",
+    "142": "SUCCESS",
     "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
-    "145": "NON_PARTICIPATING",
+    "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "SUCCESS",
-    "15": "NOT_ALIVE",
-    "150": "NON_PARTICIPATING",
-    "151": "NON_PARTICIPATING",
-    "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
+    "15": "FAIL",
+    "150": "NOT_ALIVE",
+    "151": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NON_PARTICIPATING",
-    "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "156": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "SUCCESS",
-    "159": "NON_PARTICIPATING",
-    "16": "SUCCESS",
+    "159": "SUCCESS",
+    "16": "FAIL",
     "160": "SUCCESS",
-    "161": "SUCCESS",
+    "161": "NON_PARTICIPATING",
     "162": "SUCCESS",
-    "163": "NOT_ALIVE",
+    "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
-    "165": "SUCCESS",
+    "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
-    "168": "NOT_ALIVE",
-    "169": "NOT_ALIVE",
+    "168": "SUCCESS",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "SUCCESS",
-    "171": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
-    "174": "NON_PARTICIPATING",
-    "175": "SUCCESS",
+    "174": "NOT_ALIVE",
+    "175": "FAIL",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
-    "178": "NON_PARTICIPATING",
+    "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
+    "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
+    "182": "NON_PARTICIPATING",
+    "183": "NOT_ALIVE",
     "184": "SUCCESS",
     "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
+    "187": "NOT_ALIVE",
     "188": "SUCCESS",
     "189": "FAIL",
     "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
-    "191": "NOT_ALIVE",
-    "192": "NON_PARTICIPATING",
-    "193": "SUCCESS",
-    "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
-    "196": "NON_PARTICIPATING",
+    "191": "NON_PARTICIPATING",
+    "192": "NOT_ALIVE",
+    "193": "NON_PARTICIPATING",
+    "194": "NOT_ALIVE",
+    "195": "NOT_ALIVE",
+    "196": "SUCCESS",
     "197": "NOT_ALIVE",
-    "198": "SUCCESS",
+    "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
-    "203": "NOT_ALIVE",
-    "204": "SUCCESS",
-    "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "202": "NOT_ALIVE",
+    "203": "NON_PARTICIPATING",
+    "204": "NON_PARTICIPATING",
+    "205": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
-    "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
-    "211": "SUCCESS",
+    "208": "NON_PARTICIPATING",
+    "209": "NOT_ALIVE",
+    "21": "SUCCESS",
+    "210": "SUCCESS",
+    "211": "FAIL",
     "212": "NOT_ALIVE",
-    "213": "SUCCESS",
-    "214": "NON_PARTICIPATING",
-    "215": "NON_PARTICIPATING",
+    "213": "FAIL",
+    "214": "NOT_ALIVE",
+    "215": "NOT_ALIVE",
     "216": "SUCCESS",
     "217": "SUCCESS",
     "218": "FAIL",
     "219": "SUCCESS",
     "22": "SUCCESS",
-    "220": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "NON_PARTICIPATING",
-    "224": "NON_PARTICIPATING",
+    "223": "NOT_ALIVE",
+    "224": "NOT_ALIVE",
     "225": "SUCCESS",
-    "226": "SUCCESS",
-    "227": "NOT_ALIVE",
+    "226": "NON_PARTICIPATING",
+    "227": "SUCCESS",
     "228": "SUCCESS",
-    "229": "NON_PARTICIPATING",
-    "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
+    "229": "NOT_ALIVE",
+    "23": "NON_PARTICIPATING",
+    "230": "NON_PARTICIPATING",
     "231": "NOT_ALIVE",
-    "232": "NOT_ALIVE",
-    "233": "NON_PARTICIPATING",
-    "234": "NOT_ALIVE",
-    "235": "SUCCESS",
+    "232": "FAIL",
+    "233": "NOT_ALIVE",
+    "234": "SUCCESS",
+    "235": "FAIL",
     "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
+    "240": "SUCCESS",
     "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
-    "243": "NON_PARTICIPATING",
-    "244": "NON_PARTICIPATING",
-    "245": "SUCCESS",
-    "246": "SUCCESS",
+    "242": "NOT_ALIVE",
+    "243": "NOT_ALIVE",
+    "244": "NOT_ALIVE",
+    "245": "FAIL",
+    "246": "NON_PARTICIPATING",
     "247": "SUCCESS",
     "248": "SUCCESS",
-    "249": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "SUCCESS",
+    "250": "FAIL",
     "251": "SUCCESS",
-    "252": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "SUCCESS",
-    "26": "NOT_ALIVE",
+    "26": "FAIL",
     "27": "SUCCESS",
     "28": "NON_PARTICIPATING",
     "29": "SUCCESS",
-    "3": "NOT_ALIVE",
-    "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
+    "3": "SUCCESS",
+    "30": "NON_PARTICIPATING",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
-    "33": "NON_PARTICIPATING",
-    "34": "NON_PARTICIPATING",
-    "35": "SUCCESS",
-    "36": "SUCCESS",
+    "33": "NOT_ALIVE",
+    "34": "NOT_ALIVE",
+    "35": "FAIL",
+    "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
-    "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
-    "41": "SUCCESS",
-    "42": "NOT_ALIVE",
-    "43": "SUCCESS",
-    "44": "SUCCESS",
-    "45": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
+    "4": "FAIL",
+    "40": "NOT_ALIVE",
+    "41": "NON_PARTICIPATING",
+    "42": "NON_PARTICIPATING",
+    "43": "FAIL",
+    "44": "NON_PARTICIPATING",
+    "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
-    "47": "NON_PARTICIPATING",
+    "47": "NOT_ALIVE",
     "48": "SUCCESS",
-    "49": "NON_PARTICIPATING",
+    "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "SUCCESS",
     "52": "NON_PARTICIPATING",
-    "53": "NOT_ALIVE",
-    "54": "NON_PARTICIPATING",
+    "53": "NON_PARTICIPATING",
+    "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NON_PARTICIPATING",
+    "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
-    "62": "NOT_ALIVE",
-    "63": "NON_PARTICIPATING",
+    "61": "NON_PARTICIPATING",
+    "62": "NON_PARTICIPATING",
+    "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
-    "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
-    "68": "NON_PARTICIPATING",
+    "65": "NON_PARTICIPATING",
+    "66": "NOT_ALIVE",
+    "67": "NOT_ALIVE",
+    "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
-    "70": "NOT_ALIVE",
+    "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
-    "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
-    "74": "NOT_ALIVE",
+    "72": "NOT_ALIVE",
+    "73": "NOT_ALIVE",
+    "74": "NON_PARTICIPATING",
     "75": "NOT_ALIVE",
     "76": "NON_PARTICIPATING",
-    "77": "NON_PARTICIPATING",
-    "78": "NON_PARTICIPATING",
+    "77": "NOT_ALIVE",
+    "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
-    "8": "NOT_ALIVE",
+    "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "NON_PARTICIPATING",
-    "87": "NON_PARTICIPATING",
-    "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
-    "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
+    "87": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
+    "89": "NON_PARTICIPATING",
+    "9": "NON_PARTICIPATING",
+    "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
-    "93": "NOT_ALIVE",
+    "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
-    "95": "NON_PARTICIPATING",
-    "96": "NOT_ALIVE",
-    "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
+    "95": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
+    "97": "NON_PARTICIPATING",
+    "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6420096,
-    6420112
+    6420264,
+    6420272
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 160,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6415211,
+  "last_allreduce_block": 6420196,
   "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:c5eb94fe96223bb9aa012ca40bfaf3cb55e1f78333915b207112e7b3dbdf7266
+oid sha256:fa85aae28e5d3c1f3ee58da306e69930134ce2aad7e71e38b58d577459ea43eb
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42198e61b0c019d8c0c4044cd28fa5fce9908f32539cb354cff6b908e53879f7
+oid sha256:51f5256646322ab653375ef7c5f8205dc2d15a2eb08aacb95c164b5803236d71
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f3b203df0430e558e05a23cdec4adac4fe176269e09b9f40a0c5966cd2b09b4
+oid sha256:1e97581318be8ec08060f76da561a1ace8b6738b8981f16e35cf954a0a497ede
 size 4400216536