Run 8. Outer Step 114. Inner Step 5.

config.json

@@ -3,80 +3,80 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "SUCCESS",
-    "100": "NON_PARTICIPATING",
-    "101": "NON_PARTICIPATING",
+    "10": "NOT_ALIVE",
+    "100": "NOT_ALIVE",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
-    "103": "SUCCESS",
+    "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
-    "109": "SUCCESS",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
-    "110": "NON_PARTICIPATING",
-    "111": "SUCCESS",
+    "110": "NOT_ALIVE",
+    "111": "FAIL",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "SUCCESS",
-    "118": "SUCCESS",
-    "119": "NON_PARTICIPATING",
+    "117": "FAIL",
+    "118": "NOT_ALIVE",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
-    "122": "NOT_ALIVE",
+    "121": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
-    "124": "NOT_ALIVE",
+    "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
-    "128": "NON_PARTICIPATING",
+    "127": "SUCCESS",
+    "128": "FAIL",
     "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NON_PARTICIPATING",
-    "131": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
+    "131": "NOT_ALIVE",
     "132": "NON_PARTICIPATING",
     "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
-    "137": "NON_PARTICIPATING",
+    "137": "SUCCESS",
     "138": "NON_PARTICIPATING",
-    "139": "SUCCESS",
+    "139": "FAIL",
     "14": "SUCCESS",
     "140": "NOT_ALIVE",
-    "141": "SUCCESS",
-    "142": "SUCCESS",
+    "141": "NOT_ALIVE",
+    "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
     "146": "SUCCESS",
     "147": "NON_PARTICIPATING",
-    "148": "NON_PARTICIPATING",
+    "148": "NOT_ALIVE",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NON_PARTICIPATING",
-    "151": "NON_PARTICIPATING",
-    "152": "SUCCESS",
+    "150": "NOT_ALIVE",
+    "151": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
     "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
+    "154": "SUCCESS",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "SUCCESS",
     "158": "SUCCESS",
-    "159": "NON_PARTICIPATING",
-    "16": "NOT_ALIVE",
-    "160": "SUCCESS",
+    "159": "NOT_ALIVE",
+    "16": "FAIL",
+    "160": "FAIL",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
+    "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
@@ -85,23 +85,23 @@
     "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
     "173": "SUCCESS",
-    "174": "NON_PARTICIPATING",
-    "175": "SUCCESS",
+    "174": "NOT_ALIVE",
+    "175": "FAIL",
     "176": "NOT_ALIVE",
-    "177": "SUCCESS",
+    "177": "NOT_ALIVE",
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "SUCCESS",
+    "182": "FAIL",
     "183": "SUCCESS",
-    "184": "NON_PARTICIPATING",
+    "184": "SUCCESS",
     "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
-    "189": "SUCCESS",
+    "189": "FAIL",
     "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
@@ -109,154 +109,154 @@
     "193": "NON_PARTICIPATING",
     "194": "SUCCESS",
     "195": "NON_PARTICIPATING",
-    "196": "NOT_ALIVE",
+    "196": "FAIL",
     "197": "NOT_ALIVE",
     "198": "SUCCESS",
-    "199": "FAIL",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "NOT_ALIVE",
-    "204": "NON_PARTICIPATING",
+    "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "206": "NOT_ALIVE",
     "207": "NON_PARTICIPATING",
     "208": "NON_PARTICIPATING",
-    "209": "NOT_ALIVE",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "SUCCESS",
-    "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
-    "213": "NON_PARTICIPATING",
+    "210": "FAIL",
+    "211": "FAIL",
+    "212": "NON_PARTICIPATING",
+    "213": "SUCCESS",
     "214": "SUCCESS",
     "215": "NON_PARTICIPATING",
-    "216": "SUCCESS",
-    "217": "SUCCESS",
-    "218": "SUCCESS",
-    "219": "SUCCESS",
-    "22": "SUCCESS",
-    "220": "SUCCESS",
+    "216": "FAIL",
+    "217": "NON_PARTICIPATING",
+    "218": "FAIL",
+    "219": "NON_PARTICIPATING",
+    "22": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "SUCCESS",
     "222": "NON_PARTICIPATING",
-    "223": "SUCCESS",
+    "223": "FAIL",
     "224": "NON_PARTICIPATING",
     "225": "SUCCESS",
-    "226": "NOT_ALIVE",
+    "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
-    "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "228": "SUCCESS",
+    "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
+    "230": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
-    "234": "SUCCESS",
+    "234": "FAIL",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "SUCCESS",
-    "238": "SUCCESS",
+    "237": "FAIL",
+    "238": "FAIL",
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
     "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
+    "242": "NOT_ALIVE",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "SUCCESS",
+    "245": "FAIL",
     "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
-    "248": "SUCCESS",
-    "249": "NOT_ALIVE",
+    "247": "FAIL",
+    "248": "FAIL",
+    "249": "FAIL",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "SUCCESS",
+    "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
-    "253": "NOT_ALIVE",
+    "253": "NON_PARTICIPATING",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "FAIL",
     "27": "SUCCESS",
     "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
     "3": "NOT_ALIVE",
-    "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
-    "32": "NOT_ALIVE",
+    "30": "NON_PARTICIPATING",
+    "31": "NON_PARTICIPATING",
+    "32": "NON_PARTICIPATING",
     "33": "NON_PARTICIPATING",
-    "34": "NON_PARTICIPATING",
-    "35": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
+    "35": "NOT_ALIVE",
     "36": "FAIL",
     "37": "NON_PARTICIPATING",
     "38": "NON_PARTICIPATING",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
+    "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
-    "42": "NOT_ALIVE",
-    "43": "SUCCESS",
+    "42": "NON_PARTICIPATING",
+    "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
     "45": "NON_PARTICIPATING",
-    "46": "NON_PARTICIPATING",
+    "46": "SUCCESS",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "SUCCESS",
-    "52": "SUCCESS",
-    "53": "NON_PARTICIPATING",
+    "51": "FAIL",
+    "52": "NON_PARTICIPATING",
+    "53": "FAIL",
     "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
+    "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
+    "60": "SUCCESS",
+    "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
     "63": "NON_PARTICIPATING",
-    "64": "NON_PARTICIPATING",
+    "64": "NOT_ALIVE",
     "65": "SUCCESS",
     "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
-    "68": "NON_PARTICIPATING",
+    "68": "SUCCESS",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "FAIL",
-    "71": "NON_PARTICIPATING",
-    "72": "SUCCESS",
+    "71": "FAIL",
+    "72": "FAIL",
     "73": "NON_PARTICIPATING",
     "74": "NON_PARTICIPATING",
     "75": "NON_PARTICIPATING",
-    "76": "SUCCESS",
-    "77": "NON_PARTICIPATING",
-    "78": "NOT_ALIVE",
+    "76": "NON_PARTICIPATING",
+    "77": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "SUCCESS",
+    "86": "FAIL",
     "87": "SUCCESS",
-    "88": "NON_PARTICIPATING",
+    "88": "FAIL",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "NOT_ALIVE",
-    "92": "NOT_ALIVE",
-    "93": "NOT_ALIVE",
-    "94": "SUCCESS",
+    "91": "NON_PARTICIPATING",
+    "92": "NON_PARTICIPATING",
+    "93": "NON_PARTICIPATING",
+    "94": "NON_PARTICIPATING",
     "95": "NON_PARTICIPATING",
-    "96": "SUCCESS",
+    "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
-    "98": "SUCCESS",
-    "99": "NON_PARTICIPATING"
+    "98": "NOT_ALIVE",
+    "99": "FAIL"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6432655,
-    6432670
+    6432758,
+    6432766
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 84,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6432065,
+  "last_allreduce_block": 6432723,
   "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 sha256:23611a290662529a8c29c2f9b6b325c76588dbdf368b7d3dab9e84286325e5af
+oid sha256:151bc6585d76bf456b83260dd93c1eff92f4e074740452ae732fd2e8156de929
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d946dc09fce204f70e034b4631014be8e2dbce7f4114534a15f2e34f7115e62
+oid sha256:7401870653de84e398436ae97994add56ff3e374c0a70d2b7f233bf325852ea6
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dc1fe5347b977559848beaa85e4cf75873966b52d5198196167a00d45a8e99f
+oid sha256:b4eb2d27ef20c7ec9b3ec5376fe8f2e84166c186375e41d33b1c1065a51d4ab4
 size 4400216536