Run 10. Outer Step 46. Inner Step 9.

config.json

@@ -5,12 +5,12 @@
     "1": "SUCCESS",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
-    "102": "SUCCESS",
-    "103": "NON_PARTICIPATING",
+    "101": "NON_PARTICIPATING",
+    "102": "NOT_ALIVE",
+    "103": "NOT_ALIVE",
     "104": "NOT_ALIVE",
     "105": "SUCCESS",
-    "106": "NOT_ALIVE",
+    "106": "NON_PARTICIPATING",
     "107": "NOT_ALIVE",
     "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
@@ -18,25 +18,25 @@
     "110": "NOT_ALIVE",
     "111": "SUCCESS",
     "112": "NON_PARTICIPATING",
-    "113": "NON_PARTICIPATING",
+    "113": "SUCCESS",
     "114": "SUCCESS",
     "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
-    "118": "NOT_ALIVE",
+    "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
     "12": "NON_PARTICIPATING",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
     "123": "SUCCESS",
-    "124": "NON_PARTICIPATING",
+    "124": "NOT_ALIVE",
     "125": "SUCCESS",
-    "126": "SUCCESS",
-    "127": "NOT_ALIVE",
+    "126": "NOT_ALIVE",
+    "127": "NON_PARTICIPATING",
     "128": "SUCCESS",
     "129": "NON_PARTICIPATING",
-    "13": "SUCCESS",
+    "13": "NON_PARTICIPATING",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
@@ -51,10 +51,10 @@
     "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
     "142": "SUCCESS",
-    "143": "NOT_ALIVE",
-    "144": "NON_PARTICIPATING",
+    "143": "NON_PARTICIPATING",
+    "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "SUCCESS",
+    "146": "NOT_ALIVE",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
@@ -66,23 +66,23 @@
     "154": "SUCCESS",
     "155": "NOT_ALIVE",
     "156": "SUCCESS",
-    "157": "NOT_ALIVE",
+    "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
     "159": "SUCCESS",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
-    "161": "NOT_ALIVE",
+    "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
     "163": "SUCCESS",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
+    "166": "SUCCESS",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "SUCCESS",
+    "169": "NOT_ALIVE",
     "17": "NON_PARTICIPATING",
     "170": "SUCCESS",
-    "171": "NOT_ALIVE",
+    "171": "SUCCESS",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
@@ -94,8 +94,8 @@
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
-    "182": "NOT_ALIVE",
-    "183": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
+    "183": "SUCCESS",
     "184": "SUCCESS",
     "185": "NOT_ALIVE",
     "186": "SUCCESS",
@@ -120,15 +120,15 @@
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "205": "SUCCESS",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
     "21": "SUCCESS",
-    "210": "NOT_ALIVE",
+    "210": "NON_PARTICIPATING",
     "211": "SUCCESS",
-    "212": "SUCCESS",
+    "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
     "215": "NON_PARTICIPATING",
@@ -136,19 +136,19 @@
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
     "219": "SUCCESS",
-    "22": "NON_PARTICIPATING",
+    "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "SUCCESS",
     "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
-    "226": "SUCCESS",
-    "227": "SUCCESS",
+    "226": "NON_PARTICIPATING",
+    "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
-    "230": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
     "232": "SUCCESS",
     "233": "NON_PARTICIPATING",
@@ -156,7 +156,7 @@
     "235": "SUCCESS",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
+    "238": "SUCCESS",
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
@@ -165,21 +165,21 @@
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
     "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "SUCCESS",
+    "250": "NOT_ALIVE",
     "251": "NOT_ALIVE",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "SUCCESS",
-    "255": "NON_PARTICIPATING",
-    "26": "SUCCESS",
+    "255": "SUCCESS",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
-    "29": "SUCCESS",
+    "29": "NOT_ALIVE",
     "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
     "31": "SUCCESS",
@@ -190,7 +190,7 @@
     "36": "NON_PARTICIPATING",
     "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
+    "39": "SUCCESS",
     "4": "SUCCESS",
     "40": "SUCCESS",
     "41": "NOT_ALIVE",
@@ -198,11 +198,11 @@
     "43": "NOT_ALIVE",
     "44": "SUCCESS",
     "45": "NON_PARTICIPATING",
-    "46": "NOT_ALIVE",
+    "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
-    "5": "NON_PARTICIPATING",
+    "5": "NOT_ALIVE",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
@@ -212,7 +212,7 @@
     "56": "NOT_ALIVE",
     "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "SUCCESS",
     "61": "NON_PARTICIPATING",
@@ -220,13 +220,13 @@
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
     "65": "NOT_ALIVE",
-    "66": "NON_PARTICIPATING",
+    "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
-    "7": "SUCCESS",
+    "7": "NOT_ALIVE",
     "70": "NON_PARTICIPATING",
-    "71": "NOT_ALIVE",
+    "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "SUCCESS",
@@ -235,8 +235,8 @@
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
-    "8": "NOT_ALIVE",
-    "80": "NOT_ALIVE",
+    "8": "NON_PARTICIPATING",
+    "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
@@ -245,7 +245,7 @@
     "86": "SUCCESS",
     "87": "NON_PARTICIPATING",
     "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
+    "89": "SUCCESS",
     "9": "NON_PARTICIPATING",
     "90": "NOT_ALIVE",
     "91": "NON_PARTICIPATING",
@@ -253,7 +253,7 @@
     "93": "NOT_ALIVE",
     "94": "SUCCESS",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
+    "96": "SUCCESS",
     "97": "NOT_ALIVE",
     "98": "NOT_ALIVE",
     "99": "NOT_ALIVE"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6493037,
-    6493073
+    6493221,
+    6493251
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 31,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6492801,
+  "last_allreduce_block": 6493182,
   "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:df199e63fe3943ab1a5fb3762fb19de8adbabc467a25e0b4dc14f8990acbfeb8
+oid sha256:3a02609970088616d16aa842d7cd0154d0637ec1c591af25459a2f8377ab55df
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e5855cac4eda8b7202f167c8a190059d3723d19161c71a33420b0b64d74a285
+oid sha256:703383b56961bfbfc6a9e3c6f15022c3eb877c901e02940bcf5f161a4ff85afe
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b95b8c18b031323a617efca9d7325cf615e4c7c9e6734b30754018f61fd5f7d
+oid sha256:1fef119660deadf21c11c182f8cc93235ed046161932a1a6cb36cdda85d35fb3
 size 4400216536