Run 8. Outer Step 115. Inner Step 5.

config.json

@@ -5,117 +5,117 @@
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
-    "106": "NOT_ALIVE",
+    "106": "SUCCESS",
     "107": "NON_PARTICIPATING",
-    "108": "NON_PARTICIPATING",
-    "109": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
+    "109": "FAIL",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
-    "111": "FAIL",
+    "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
-    "113": "NON_PARTICIPATING",
+    "113": "SUCCESS",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "FAIL",
+    "117": "NON_PARTICIPATING",
     "118": "NOT_ALIVE",
-    "119": "NOT_ALIVE",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
-    "125": "NOT_ALIVE",
+    "125": "SUCCESS",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "FAIL",
-    "129": "NON_PARTICIPATING",
-    "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
-    "131": "NOT_ALIVE",
-    "132": "NON_PARTICIPATING",
+    "128": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
+    "13": "SUCCESS",
+    "130": "NON_PARTICIPATING",
+    "131": "NON_PARTICIPATING",
+    "132": "SUCCESS",
     "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
     "137": "SUCCESS",
     "138": "NON_PARTICIPATING",
-    "139": "FAIL",
-    "14": "SUCCESS",
+    "139": "NON_PARTICIPATING",
+    "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "SUCCESS",
+    "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "NOT_ALIVE",
-    "149": "NOT_ALIVE",
+    "148": "NON_PARTICIPATING",
+    "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
-    "151": "NOT_ALIVE",
-    "152": "NON_PARTICIPATING",
+    "150": "NON_PARTICIPATING",
+    "151": "NON_PARTICIPATING",
+    "152": "SUCCESS",
     "153": "NON_PARTICIPATING",
-    "154": "SUCCESS",
+    "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "SUCCESS",
     "158": "SUCCESS",
     "159": "NOT_ALIVE",
-    "16": "FAIL",
-    "160": "FAIL",
+    "16": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "SUCCESS",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
-    "171": "NON_PARTICIPATING",
+    "171": "SUCCESS",
     "172": "NON_PARTICIPATING",
-    "173": "SUCCESS",
-    "174": "NOT_ALIVE",
-    "175": "FAIL",
+    "173": "NOT_ALIVE",
+    "174": "NON_PARTICIPATING",
+    "175": "SUCCESS",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "NON_PARTICIPATING",
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "FAIL",
+    "182": "SUCCESS",
     "183": "SUCCESS",
     "184": "SUCCESS",
-    "185": "NON_PARTICIPATING",
+    "185": "SUCCESS",
     "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
-    "189": "FAIL",
+    "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "190": "SUCCESS",
     "191": "NOT_ALIVE",
-    "192": "NON_PARTICIPATING",
+    "192": "NOT_ALIVE",
     "193": "NON_PARTICIPATING",
     "194": "SUCCESS",
-    "195": "NON_PARTICIPATING",
-    "196": "FAIL",
+    "195": "NOT_ALIVE",
+    "196": "SUCCESS",
     "197": "NOT_ALIVE",
-    "198": "SUCCESS",
+    "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "NON_PARTICIPATING",
+    "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "NOT_ALIVE",
@@ -124,27 +124,27 @@
     "206": "NOT_ALIVE",
     "207": "NON_PARTICIPATING",
     "208": "NON_PARTICIPATING",
-    "209": "NON_PARTICIPATING",
+    "209": "SUCCESS",
     "21": "NOT_ALIVE",
-    "210": "FAIL",
-    "211": "FAIL",
+    "210": "SUCCESS",
+    "211": "SUCCESS",
     "212": "NON_PARTICIPATING",
     "213": "SUCCESS",
-    "214": "SUCCESS",
+    "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "FAIL",
-    "217": "NON_PARTICIPATING",
-    "218": "FAIL",
-    "219": "NON_PARTICIPATING",
+    "216": "SUCCESS",
+    "217": "NOT_ALIVE",
+    "218": "SUCCESS",
+    "219": "SUCCESS",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "NOT_ALIVE",
     "221": "SUCCESS",
-    "222": "NON_PARTICIPATING",
-    "223": "FAIL",
+    "222": "SUCCESS",
+    "223": "NON_PARTICIPATING",
     "224": "NON_PARTICIPATING",
-    "225": "SUCCESS",
-    "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "225": "NON_PARTICIPATING",
+    "226": "SUCCESS",
+    "227": "SUCCESS",
     "228": "SUCCESS",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
@@ -152,11 +152,11 @@
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
-    "234": "FAIL",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "FAIL",
-    "238": "FAIL",
+    "237": "NON_PARTICIPATING",
+    "238": "SUCCESS",
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
@@ -164,51 +164,51 @@
     "242": "NOT_ALIVE",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "FAIL",
+    "245": "NOT_ALIVE",
     "246": "NON_PARTICIPATING",
-    "247": "FAIL",
-    "248": "FAIL",
-    "249": "FAIL",
+    "247": "SUCCESS",
+    "248": "SUCCESS",
+    "249": "SUCCESS",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
-    "251": "NON_PARTICIPATING",
+    "250": "NOT_ALIVE",
+    "251": "SUCCESS",
     "252": "NON_PARTICIPATING",
-    "253": "NON_PARTICIPATING",
+    "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "FAIL",
+    "26": "NON_PARTICIPATING",
     "27": "SUCCESS",
     "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
     "3": "NOT_ALIVE",
     "30": "NON_PARTICIPATING",
-    "31": "NON_PARTICIPATING",
+    "31": "NOT_ALIVE",
     "32": "NON_PARTICIPATING",
     "33": "NON_PARTICIPATING",
-    "34": "NOT_ALIVE",
+    "34": "NON_PARTICIPATING",
     "35": "NOT_ALIVE",
     "36": "FAIL",
-    "37": "NON_PARTICIPATING",
+    "37": "NOT_ALIVE",
     "38": "NON_PARTICIPATING",
-    "39": "NON_PARTICIPATING",
-    "4": "NOT_ALIVE",
+    "39": "NOT_ALIVE",
+    "4": "SUCCESS",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
-    "43": "NOT_ALIVE",
+    "43": "SUCCESS",
     "44": "NON_PARTICIPATING",
     "45": "NON_PARTICIPATING",
     "46": "SUCCESS",
-    "47": "NON_PARTICIPATING",
+    "47": "NOT_ALIVE",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "FAIL",
-    "52": "NON_PARTICIPATING",
-    "53": "FAIL",
+    "51": "NON_PARTICIPATING",
+    "52": "SUCCESS",
+    "53": "NON_PARTICIPATING",
     "54": "NON_PARTICIPATING",
-    "55": "NOT_ALIVE",
+    "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
@@ -217,46 +217,46 @@
     "60": "SUCCESS",
     "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
-    "63": "NON_PARTICIPATING",
-    "64": "NOT_ALIVE",
+    "63": "NOT_ALIVE",
+    "64": "NON_PARTICIPATING",
     "65": "SUCCESS",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
-    "70": "FAIL",
-    "71": "FAIL",
-    "72": "FAIL",
+    "70": "NON_PARTICIPATING",
+    "71": "NON_PARTICIPATING",
+    "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
-    "74": "NON_PARTICIPATING",
+    "74": "SUCCESS",
     "75": "NON_PARTICIPATING",
-    "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
-    "78": "NON_PARTICIPATING",
+    "76": "SUCCESS",
+    "77": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
-    "81": "NON_PARTICIPATING",
-    "82": "NOT_ALIVE",
+    "81": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
-    "87": "SUCCESS",
-    "88": "FAIL",
+    "86": "NOT_ALIVE",
+    "87": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
-    "92": "NON_PARTICIPATING",
+    "91": "SUCCESS",
+    "92": "SUCCESS",
     "93": "NON_PARTICIPATING",
-    "94": "NON_PARTICIPATING",
+    "94": "SUCCESS",
     "95": "NON_PARTICIPATING",
     "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
     "98": "NOT_ALIVE",
-    "99": "FAIL"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6433282,
-    6433296
+    6433414,
+    6433432
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 86,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6432723,
+  "last_allreduce_block": 6433371,
   "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:f8631add1eb7740400e986c66249bbfee5c842df594ae4e79fd49a59409a4f8c
+oid sha256:541d2b4824ebbe17590abcd0e25bdac2f927338677911cb84b8f7d03ea74ff23
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12f644e4c5420a293b5fd146a72492c0ed0145eeb9b991dd9047d44c586aea79
+oid sha256:baaba77ccc66a9996ede056159dd5ec9c93c000f4df4106376ba0f685494b64d
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2cc5db42b86a0a447ade5620113d288a8e1d2256d419b2a41362247b08092d5
+oid sha256:ae3752b261a9bbe87174549318b5f6d4ac2b62868dd595295b8698975ee03183
 size 4400216536